WO2014187432A1 - Video iv monitor - Google Patents

Abstract

This application is about image processing based monitoring of IV administration. The disclosure primarily includes 1) Heating methods for removing splash and condensation droplets 2) Optical (lens) system design and the analysis of the PVC drip chamber's effect upon droplets' volume measurement, and the associated measures for accuracy improvement 3) The correction of volume measurement due to distance, height, rotational and tilting angle variations, and use of Laplace-Young equation to fit the droplet image profile 4) The use of anti-splash drip chamber together with the video monitor.

Description

Video IV Monitor

Inventor: Kai Tao

Medical Infusion Devices

Background ~ prior art

The purpose of this invention is to create a device that could achieve accurate volumetric measurement of falling droplets during an IV administration, and it works by video and image processing techniques. It intends to compete with existing mechanical infusion pumps with advantages in cost, precision and ease of use.

This is a relatively new field. Except from applications from the present applicant himself, prior arts include:

CN 200710168672.3, Enmin Song, A medical infusion speed monitoring and controlling system, which contains generic description of a system also based on image processing. This prior art doesn't touch the specific problems in our present application.

US12791885, Ting- Yuan Cheng, "Intravenous Drip Monitoring Method And Related Intravenous Drip Monitoring System", which also contains only generic descriptions of a very rudimentary system based on the same principle.

The description is divided into five parts denoted by letter A, B, C, D and E, and the associated drawings are also numbered with these letters.

Part A incorporates primarily the content of PCT/IB2013/056090 and discusses condensation removal and optical correction. Part B continues on A and discusses further methods of condensation/splash prevention/treatments. Part C continues on optical and tilt correction. Part D uses mechanical vibration/resonance to remove splash/condensation. Part E discusses calibration library and dust removal.

This application is so organized such that the respective parts in appearance maintains their original structure, particularly in that each part may have its separate introduction and some preliminary discussion. However, abstract, claims and drawings are certainly placed to together to conform with formality requirements, and when words like "description" "introduction" "brief description of drawings" "summary" appear in text they should not be construed to disturb the integrity of the main and regular "description" part. Summary

In this application we describes some essential features which makes an image/video processing based IV monitoring and controlling devices comparable or could achieve better accuracy than traditional mechanical infusion pumps. Clear images is obviously a prerequisite for devices working under this principle, however the existence of condensation and splash droplets on the inner surface of the drip chamber could prevent the camera from seeing the droplets clearly and leads to inaccuracies in volume measurements. Therefore we first describe various ways of removing these droplets either by direct or indirect heating (radiation). The problems are complicated because of the thermal and radiation absorption characteristics of IV set materials, and the heating methods' applicability must be rigorously established by theoretical calculations.

After that we will discuss methods for obtaining high-quality droplets images in the absence of condensation/splash droplets. Some illumination methods will be described. Then we will show how the plastic materials of the IV sets themselves will behave as a lens and have an effect upon the real optical system, and how will different parameters of the drip chamber influence droplet volumetric measurement accuracy. Methods and apparatus for obtaining the drip chamber parameters and for correcting their effects are disclosed.

We also describe how a separate bedside monitor could be used in conjunction with the display-equipped "master" device to ease to allow patient monitor and operate the device conveniently.

In the end, we combine the how splash/condensation droplet removal heating, a non-transparent full-enclosure, and accelerometer could be used to improve the accuracy of infrared drop counters.

Brief Description of the Drawings

Fig.A.1-1 to Fig.A.1-4 shows splash and condensation droplets on drip chamber.

Fig. A.1.1 shows the use of contact heating (conduction) to remove splash and condensation droplets.

Fig.A.1.2.1-(l to 2) shows the simulation model used for a 3.8μπι infrared emitter heater.

Fig. A.1.2.1 -(3 to 12) shows the absorption spectrum for saturated water vapor mixture with air components over range [0.497512, 99.9989] μπι.

Fig.A.1.2.2-(l to 7) shows FTIR measured absorption spectrum for our five PVC plastic IV drip chamber samples. Fig.A.1.2.3-(l to 2) shows the power spectrum of a 3.8μπι infrared emitter.

Fig.A.1.2.5-(l to 2) shows the PVC chamber's absorption of the infrared emitter's spectrum and the remaining. Fig.A.1.3.1-(l to 5) shows the experiment for incandescent light heating. Fig.A.1.3.3-(l to 2) illustrates the PVC and glass transmission windows.

Fig.A.1.3.4-(l to 6) shows calculations and simulations results for incandescent light heating.

Fig.A.1.3-5-(l to 4) illustrate our inventive elements for optimal illumination.

Fig.A.1.3.5-5(1) shows a TES-1310 thermometer used in our incandescent bulb experiments.

Fig. A.1.3.5-5(2 & 3) shows a constant temperature incubator and a constant temperature water bath used for studying IV drip chamber's condensation phenomenon.

Fig. A.2.1 illustrates the "blocking effects" of droplets.

Fig. A.2.2 illustrates the "cylindrical lens" formed by PVC drip chamber and its modeling in Zemax. Fig.A.2.3 shows a lens system.

Fig. A.2.4-(l to 7) show our simulation in Zemax for varying combinations of parameters, and post-analysis data.

Fig.A.2.6-(l to 4) shows our inventive elements pertaining to the optical system.

Fig. A.3-1 illustrates a bedside remote monitor/controller for our IV monitoring device.

Fig. A.4 illustrates our improvements to the traditional infrared drop counter.

.4.1. Heating for condensation and splash droplets removal

Please refer to Fig.A.1.1 to 1.4. Fig.A.1-4 was taken during the normal course of an IV dripping experiment and the liquid was just tap water. There are from several dozens to up a hundred small droplets attaching on the vertical inner surface of the IV tube. Fig.A.1-1 and 1-2 illustrated their effect of obscuring the view of a falling drip, whose size (and hence volume) and location of our primary interest. When these condensation/splash droplets are really severe, no image processing/enhancement could be able to guarantee the accuracy of volumetric measurement of droplets.

These are real, not made-up effects and can be observed in the IV clinic room of all hospitals, and they happen on virtually all brands of IV sets. They happen because

1. The drip chamber is a closed environment and there is no way for the constantly evaporated water vapor to escape this environment, so very quickly the chamber is filled with saturated water vapor.

2. The mere existence of saturated water vapor would not cause surface droplets to form quickly. If there is no dripping, the saturated vapor takes dozens of minutes or even hours to form a thin layer of tiny fog on the surface and there are much fewer large, discernable droplets as can be seen in Fig.A.1 -4. This is because the formation of condensation droplets requires condensation nuclei, like the formation of clouds is facilitated when the air contain a large number of small particles (including pollutants). The dropwise condensation process have been studied with imaging, microscopy and computer simulation methods (R. N. Leach, Dropwise condensation: Experiments and simulations of nucleation and growth of water drops in a cooling system , and other papers), and was found to be essentially consisting of saturated vapor first forming very small droplets, and these small droplets will then coalesce to grow larger and larger and eventually become effective condensation nuclei to attract more vapor condensation. At room conditions when the drop chamber surface's temperature is very close to the temperature of the vapor, this is a very slow process. However when the dripping is on, the splash droplets caused by the falling drips would effectively behave like initial condensation nuclei, and they will in turn cause more saturated water vapor in the chamber to condense, so in a short time a large number of splash and condensation droplets will form on the chamber's inner surface.

3. Drug could complicate the process. Chemicals with low surface energy, include various waxes and fatty acids such as oleic, stearic, and linoic acids have been used in industry as "promoters" to facilitate and sustain dropwise condensation (Cengel - Heat Transfer: A Practical Approach, section 10.7), and there are many types of drugs with similar chemical characteristics so they could effectively act as "promoters" and will cause more significant droplets accumulation.

The general principle to prevent these droplets' formation and remove them after they have formed is to heat the drip chamber surface. If the temperature of an area of the surface could be maintained at sufficiently higher than the internal gas mixture temperature, then 1) water molecules in the vapor themselves would not spontaneously condense on the surface because when molecules get in contact with the surface, the higher kinetic energy the surface imparts to them only makes the vapor thermodynamically more stable in the gas region [Atkins - Physical Chemistry (8ed), chapter 4]; 2) the splash droplets will soon be evaporated, so gaseous vapor would have very little chance of condensing on them and then coalesce into large droplets.

There are three mechanisms of heat transfer: conduction, convection and radiation. We will show how examples in the conduction and radiation category. Our simulation has shown that in the enclosed environment of a compact medical device, it is impossible heat a specific area of the drip chamber purely air convection, and the major role of any "convective" drip chamber heating system is actually always the radiation of the heating element, therefore we only discuss conduction and radiation heating in the application.

AX.1. Heating by eon duetto®

Directing conduction heating work by arranging the heating element in direct contact with the drip chamber. This is illustrated in Fig. A.1.1-1 to 1.1-3 (also discussed extensively in App. US 13897578 and US 13903924 by the same inventor) . Since we only need to ensure a clear view for the region that the camera will look at, there is no need to heat the entire outside surface. The heater in 1.1-4 assumes the shape of abended hollow rectangle, having the inner(back) side surface in contacting (hugging) the chamber directly. The hollow region in the middle is for exposing the view of the droplet to the camera.

The material of the heater is preferably copper for two reasons: 1) copper has a thermal conductivity of 398 W m ' -K^"1, higher than most metals and other materials 2) its emissivity is only 0.03 (for electroplated copper), so very little heat is converted as thermal radiation. The faces of the heater not facing the drip chamber should in general be wrapped by thermal insulators to avoid energy waste, and these insulators can be heat insulating paints, coatings, or simply poor thermal conductors like plastic, and this is clearly for the purpose of avoiding unnecessary heat dissipation through air convection.

The source of heating energy is should be considered a separate from the form of heating. Most conveniently one could use Joule (resistive) heating or thermoelectric effect (Peltier/Seebeck effects, etc.) since these heat sources could be made very compact, however there are still other ways of generating or collecting heat such as gathering heat dissipated by the battery orPCB assembly, or even use some radioactive element (Plutonium-238 has been successfully used on artificial heat paemakers for sustained service life). The separation of the source of heating energy and the form of heating should easily be recognized, and in the remaining of the application we only discuss heating methods (forms of heating) rather than how to generate the energy.

Discussions in this patent application are quantitative. We do not merely state "how" an apparatus/method could be used to solve a problem, but also gives quantitative details on its material composition, power rating and geometric dimensions and performance. For each heating methods, we use Solidworks 2012 Flow Simulation to numerically simulate the flow and heat dynamics.

It should be noted that the results obtained by Solidworks Flow Simulation are accurate. The [Solidworks Flow Simulation 2012 technical reference] contain validation many examples comparing simulation result with classical analytical solutions and the results agree well. There are other CFD (computational fluid dynamics) software more widely used in chemistry study (such as Fluent, Phonetics) with many more specific options, but for our purpose of simulating the heat transfer near a pressure of 1 atm without chemical (molecular composition) change, the capability of Solidworks Flow Simulation is far more than enough.

Please also note that Solidworks Flow Simulation not only simulates fluid dynamics and heat transfer in liquid/gas, but also heat transfer in solids. This is why in our simulations, a heater contacting the outside surface of the plastic drip chamber could also cause the temperature of the air inside the chamber to rise.

Most drip chambers are made of PVC (Polyvinyl chloride) plastic material containing various amounts of plasticizers. The material properties {density, specific heat, etc.} are from Table. A. 12.7 of [Wilkes, Summers, Daniels, PVC Handbook] and the thermal conductivity value was from another source:

The specific heat value varies with the amount plasticizers and the Wilkes book listed the values for both pure P VC and those containing 50% DOP (Di-n-octyl phthalate). Most PVC drip chambers (and indeed the ones used in our experiment) used in China actually contain 20% to 40% DEHP (Di-2-Ethylhexyl Phthalate) as plasticizer, and the 1.5 J/(g °C) value was obtained from a manufacturer.

Air

The air was assumed to be standard composition atmospheric air at near latm (the actual air pressure in the drip chamber = external pressure + Pwater

fwaⁱer.oQh ~ l ⁴O⁰O^cO^mcm^ww^ata^et^rer i_atm = 0.04atm » 1.04 atm because of the heig ^bht of thin water cylinder above the tube, and both latm and 1 04atm can be very well approximated by ideal gas properties), and the humidity was assumed to be 100% (enclosed environment, please refer to previous discussion).

Water

In Fig.A.1.1 -4, the bottom part (show in translucent blue color) was specified as fluid water whose properties are supplied by the software.

Liquid-Gas Interface

CFD (computational fluid dynamics) software normally have difficulties in modeling Liquid-Gas Interface and works more comfortably with liquid-solid/gas-solid boundary conditions, so we model the water-air interface as a very thin sheet of metal with extremely low specific heat (0.01 J/(kg K)) and extremely high thermal conductivity (800 W/(m*K)). This thin sheet should allow heat to exchange virtually transparently across the sheet between water and air. We have consulted Solidworks experts and confirmed the validity of this approach.

Heater

The heater has a bended rectangular shape, having a thickness of 1mm. The exposed hollow window for the camera's view has height 12.53mm and width 9.63mm. Drips rarely grow over 3mm in diameter even drugs solutions with low surface tension, so this window is large enough for us to observe the falling trajectory and size of individual droplets. Please also refer to out [App. US12804163 IV Monitoring by Video and Image Processing] for why only a small observation area (window) is enough for measuring the volume and fallings speeds of the droplets.

As we have discussed above, to avoid unnecessary heat dissipation through air convection the external surfaces of the heater should be wrapped in thermal insulators. This is modeled by specifying the Heat Transfer Coefficient of the all the other walls excluding the one facing the drip chamber, to be 0 W/(m². K).

The heating power was specified as a surface heat source of 0.03M/ heat generation rate.

The steady-state results are shown using cut plot on the middle symmetric surface in Fig.A.1.1.1-1 and surface plots for the inner surface of the drip chamber in Fig. A.1.1.1-2, both contour and isolines are shown. For the middle cut plot, it is clearly that the highest temperatures are around 320.93K all at the back surface of the heater, 22K higher than ambient air (298K). The center of the PVC surface at of the interior of the heating window reaches 316.4K, more than 18K higher than the ambient air (298K) temperature. For the air temperature, in the middle of the drip chamber far from the heated surface, the temperature is only 302.52K, 4.52K higher than ambient air; when the position gets lower, the air temperature become very close to the water temperature (298.33K).

We could draw two conclusions from the cut plot:

1. Because the heater is in contact with the upper part of the drip chamber, heated air (lower density) will remain in the upper part and convection within the chamber will be very weak. Because the air temperature near the water surface remains the lowest, there will only be very little increase in the evaporation rate at the surface.

2. The most fundamental relation is the order of heating

heater→ PVC (first)→ Air (secondary)

so PVC is always heated first, and air the second. The secondary heated part can never reach a temperature higher than the first heated part. This relation guarantees that condensation will never form on the heated PVC area.

3. 320K temperature at the inner surface would cause significant evaporation. 320K = 47°C, and a

temperature of 46°C already causes pain on the skin [Cengel, Heat Tansfer, page 45]. At around 47°C (in fact also the center area 43 °C) the splash droplets would soon evaporate and will never got the chance to grow and coalesce into large dropwise condensation.

The conclusion that a direct contact heater could prevent and eliminate splash/condensation droplets should be evident, both from the first→secondary fundamental relationship, and from the simulation results.

The surface plot contains distribution of temperature around the inner cylindrical surface. It is also quite clear that due to the low thermal conductivity of PVC material, only the area directly heated increases the temperature significantly. The changes on lateral and opposite sides are weak.

Justification: why the system is enclosed

In the modeling of direct heating, we didn't not model the inlet (incoming) and outlet (outgoing) flow of liquid (dripping) (and possible some gas contained in the liquid). If we do have incoming flow, it would actually have a strong cooling effect for the internal gas mixture considering the large specific of water (even for the smaller drips). Under ordinary dripping speed 60 drips/mL, each minute there will be a net volume of 3mL to 4mL exchange of liquid (water), and they could bring away significant amount of heat from the air, so the internal air temperature could only be lower than what we have modeled for an enclosed temperature. The cooling effect of liquid flow would have a primary influence on the internal gas mixture first, and only a secondary influence on the temperature (via cooled gas, via conduction at the lower part of the drip chamber, but the thermal conductivity of PVC is poor) of the heated area of the drip chamber. In general, the air temperature in our simulation could only be higher than in actuality when there is a continuous liquid flow. The direct contact heating method has two primary drawbacks:

1) It is difficult for one heater to have seamless contact with all types of drip chambers because drip chambers can vary in external diameters. In addition, some drip chambers have a tapering cylindrical shape (the diameter of the cylinder decrease from top to bottom. The effective contacting area of the heater with the drip chamber will drastically decrease in these cases.

2) Secondly, even if we could force a seamless contact by exerting a pushing force on the direct heater (from the top view it has a meniscus shape) so that cylindrical drip chambers (not the tapering type) of a different radius than the radius of curvature of the heater would DEFORM to accommodate its shape, we are not merely deforming plastic, but also the Lens! As will be discussed in greater detail in the "Optical Design" part later in this application, the PVC drip chamber has refractive index between 1.52 and 1.54, a non- negligible thickness between 0.5mm and 1mm, and a spherical radius small in the horizontal plane but increases as we imagine the cutting sections (Fig.A.2.2-1) from horizontal to vertical plane. They effectively behave as optical lens and their effect MUST be included in the optical system. It is possible for our monitoring device to obtain the exact optical parameters {drip chamber inner and outer radius→thickness, refractive index _; } by reading some tags (barcode, etc.) or via user input, but if the contacting heater causes the drip chamber to deform, the particularly the inner and outer radius will be altered, and could adversely affect the accuracy of our system even if we do optical correction (either by using a micro motor to change Distance [object, lens]) or Distance [image, lens] or digital image correction (on the pixel level).

These drawbacks can be avoided if we could achieve the same heating effect without contacting the drip chamber.

We describe below the use of infrared radiation to heat the PVC chamber. However a major difference between infrared heating and direct contact heating is that the First→ Secondary relation is lost. The volume of the air is very small (both d and h below are measured on typical IV sets, h is the distance from water surface to top of the drip chamber): d = 1.4 cm; h = 4 cm

^air = π(- ^d)²Η = 6.1575 mL ¾ 28_fl/mol _{χ 1575mL} _ 0.0076969 g = 7.7 mg , Avogadro's law

a^lr 22.4L/mol and contains saturated water vapor, and it is well known that water vapor is a major greenhouse gas in the atmosphere which absorbs some bands of the infrared radiation strongly. In addition, the air in the drip chamber also contains C0₂ whose concentration is low but effect could also be significant in certain absorption bands.

Given the small volume and mass of gas and its high concentration of water vapor, a major IR absorber, we need to address the following questions before applying IR heating to the drip chamber: 1. The PVC layer is thin (~0.5m), so how much IR could it absorb? Could it get enough IR energy be effectively heated?

2. How much IR energy will pass the PVC and get absorbed by the gas mixture (chiefly due to H₂0 and C0₂)? The drip chamber contains saturated water vapor which is a strong radiation absorber, how many degrees will its (and the total air mixture) temperature rise?

3. Could gas mixture temperature rise to higher than chamber surface? If this happens, then the high- temperature gas would facilitate water evaporation, and the evaporated vapor will later condense on the relatively cooler drip chamber surface. In this way, we are ENCOURAGING condensation rather than preventing it!

The basic question is who will be heated to higher temperature, gas or drip chamber? There is no way to give a back- of-the-envelope calculation to get a definite result. To obtain quantitative results we must have quantitative data.

The absorption spectrum of water is very complicated due to the co-existence of vibrational, rotational resonance and electron excitation mode absorption, and is further complicated by the existence of C02, N2, 02. The definitive source of gas mixture absorption data is HITRAN (High Resolution Transmission, htt : //ww . cfi i . harvard . sdu/ht tratV) . Our data was obtained from SpectralCalc.com (paid service), calculated by the LINEPAK algorithm [L L. Gordley - LINEPAK: Algorithms for Modeling Spectral Transmittance and Radiance] based on HITRAN 2008 data. We gathered full-resolution gas mixture data with the following composition:

# Cell Number: 1

# Pressure (mb) = 1013.25

# Temperature (K) = 296

# Length (cm) = 1.4

# Gas Isotope Id Volume Mixing Ratio

# H20 0 All Isotopes 0.0507179

# C02 0 All Isotopes 0.000376851

# 02 0 All Isotopes 0.198829

# N2 0 All Isotopes 0.74121

The length was specified as 1.4cm because this is the typical inner diameter of the drip chamber. An illustration of "gas cell" is shown in Fig. A.1.2.1-1. Please also refer to the left of Fig. A.1.2.1-2: the beam from the IR emitter can be directed to travel in straight direction so that it irradiate the drip chamber in its normal direction. If the beam is not so wide, it is reasonable to ignore refraction due to the PVC material, and the majority of beam's energy will traverse what is very close to the a distance of 1.4cm, the inner diameter of the drip chamber, before it comes touch the other side of drip chamber.

The H20 volume mixing ratio of around 5% could only be reached when the air temperature is around 40°C, and this is certainly NOT the normal hospital ward/clinic temperature. It was intentionally chosen as a "safety" measure to ensure that our later calculation would always result in stronger heating (absorption due to H20) than in the normal environment. The data were calculated in the range [0.497512, 99.9989]μπι, from within visible spectrum to LWIR (long wave infrared). The total size of TXT data files is 175 MB and consisting of 4172592 distinct spectral lines due to the four mixing gas components.

The resolution of the spectral lines, according to the SpectralCalc.com, was "chosen to be fine enough to fully capture the line shape of the thinnest absorption line in the spectrum" and "set to 20% of the thinnest line's halfwidth".

The spectral line plots for different ranges are listed in Fig. A.1.2.1-3 to 1.2.1-10. Please note that the lines are essentially discrete, and in some regions they appear continuous was only due to the artifacts of graphics (e.g., if 500,000 lines (vertical) are drawn in a 1000 pixel wide area, then each line in theory could only occupy 1/500 pixie's width. Graphics software enlarge the widths so that they could be seen, but also make the spectrum appearing continuous.)

A.IJ-2 A s*q?tsos5 s ecinsssi of PV£

One absorption spectrum of pure PVC (without plasticizer) can be found in [Wilkes, PVC Handbook, Fig.A.15.40]. However, in reality all IV set drip chambers uses plasticized PVC, and the plasticizer used in over 95% of IV sets in many developing countries is still DEHP (Di-2-Ethylhexyl Phthalate). The actual transmission/absorption varies with DEHP concentration and its thickness due to Beer-Lambert Law:

/ = e^~al

In order to get accurate PVC -DEHP absorption data we have taken five samples (see Fig. A.1.2.2-1) to three material testing centers in Jiangsu province:

*FTIR stands for Fourier transform infrared spectroscopy

All values measured are absolute transmission values rather than relative intensity commonly seen in ATR (Attenuated total reflectance) values used for analyzing material compositions. The dependence on thickness can be seen clearly in FIG.A.1.2.2-2 in which we have asked the technician to test the overlay of one, two and three layers of PVC for different samples. This figure [FIG.A.1.2.2-2] was the original figure plotted by the Cary 670-FTIR machine's software, and when there is only one layer of PVC, three different samples all show have similar transmittance near 40% at 2.5um (wavenumber 4000 cm^"1); when two or three layers are overlaid, the transmittance drops. The larger discrepancy between the lower lines were due to the fact that we when we overlaid two or three layers of PVC, it was difficult to align and press them flat as perfect as a single layer, so there are some deviation between samples. Single layer PVC samples from different manufacturers have almost the same absorption spectrum.

Fig. A.1.2.2-6 from the Nanjing University testing center show the overlay of three bands plotted on single graph with different colors {blue, purple, red} with wavenumber (cm ¹) coordinate, and Fig.A.1.2.2-7 show the same graph using wavelength coordinate. Because in order to measure different wave band different accessories of the FTIR machine are used, the fact that they can be connected (lines in different color overlays well at the connection) indicate that our measurement were successful and accurate, and the values is qualified to be used in calculations.

In IR radiation heating, the most common method is to heat an object to a temperature so that its peak emission band matches the peak absorption band of the material being heated. The theoretical basis is Wien 's displacement law :

(XT)_maxpower = 2897.8|im. /f

If we look closely at Fig.A.1.2.2-4(2) (and equivalent λ plot Fig.A.1.2.2-4(1)), we notice two peak absorption (zero transmission) bands: near 3.4um and near 5.8μπι. The 3^m peak absorption was due to PVC itself whereas the 5.8μπι band was due to plasticizer (be it DEHP, sulphonate phenyl ester, terephthalate ester, citrate ester or adipate ester. These plasticizers have basically the same resonance band (peak value)). The infrared absorption of plasticizer at its band is stronger than that of PVC so for very thin layers of PVC film, the 5.8μπι absorption band dominates. Our samples (drip chamber thickness) are of thickness about 0.5mm, and for recall the Beer-Lambert formula above

I = 0.5mm is already what is considered to be "optically thick" for the two bands, so both the PVC and plasticizer could nearly 100% absorb the light. Therefore we should tune our IR source to emit peak energy at either these sources. For IR source based on heating, preferably they should be heated to either

2897.8

_{3 4} = 852.29 K = 579.14°C or

2897.8

— = 499.62 K = 226.47°C

5.8

It was not difficult to heat elements, for example using resistive heating, to this temperature. And if we indeed use heating to generate infrared ray, to protect other parts of the device and the PCB assembly from the high temperature we could enclose the heating element in an enclosure like a glass blub, and the bulb will confine the hot filament and air (if not vacuum) inside it. The bulb might be further coated with reflective coating, or simply made of band -pass IR glass in that specific heating band, so that much of the energy outside that band will be reflected back to increase the temperature of the heating element (filament, etc.). What we are describing here is in fact in principle equivalent to an IR heating bulb. Glass typically have melting point between 1400°C and 1700°C, and the low temperature the above figures suggested can well be contained in almost all types of glass bulbs even without special treatment. As we will show clear below in the discussion of a 3 8μπι IR emitter and incandescent light heating, the key for radiation heating's success is to make sure PVC drip chamber can always be heated to higher temperature than the internal gas mixture. Whatever heating band is chosen, the establishment of the technology requires calculation based on the specific absorption spectrum of the PVC drip chamber, and methods in this category would always fall in the spectrum defined by our two examples (3.8μπι IR heater and incandescent heater) below.

Heating is not the only way of generating IR radiation. Infrared emitters for the near IR band are particularly popular because they are not absorbed by atmospheric air, so they are widely used for object detection including the detection of IV falling drips. In the mid -wave (MWIR) and long-wave (LWIR) IR range, techniques have also long existed for making IR emitters, but their use are narrower and typically on in chemical analysis or military. We collected the following list of MWIR and LWIR publications from as early as 1997.

Matveev, IOFFE Physico- In(Ga)As- and InAs(Sb)- 3.2-3.4 Yes 2002 5th etc. Technical Based Hetero structure International

Institute, Russian LEDs and Detectors for Conference on Academy of the 3-5 um Spectral Mid- Sciences Range Optoelectronic s Materials and Devices

Malyuten Lashkaryov Mid-infrared LEDs 3.4, 3.6, Yes 2006 Proc. of SPIE ko & Institute of Versus Thermal Emitters 3.8, 4.2 Vol. 6368 Zinovchu Semiconductor in IR Dynamic Scene

k Physics, Ukraine Simulation Devices

In the following we prove the use of a device similar in characteristics to Lancaster university's 3.8um H2S detector (as an illustration example rather than limitation) could serve as an ideal heater for our purpose.

A J ,2,4 SR. Sisssiites- esaergy disiriinite

The IR emitting LED as disclosed in their paper has emitting spectrum as shown in Fig. A.1.2.3 - 1 , and has an output power of 8mW. Compare it with Fig. A.1.2.1-11 of the water spectrum between 2.439 and 4.7619um, we find that the gas mixture (of H20, C02, N2, 02) has strong absorption band between 2.5 and 2.8μπι, however the H2S emitter paper figure stops abruptly at 3 um, but from the trend of its change, if we interpolate the graph from 3 down to 2.8μπι it would most likely be non-zero.

What we did was manually extrapolate the Fig. A.1.2.3-1 to 2.8μπι with higher values than what an ordinary extrapolation would result in. This is also for a "safety measure" as to intentionally give gas mixture absorption more "weight" so that our calculated gas temperature rise will be higher than in reality. The extrapolated version of Fig. A.1.2.3-1 was shown in Fig. A.1.2.3-2, and the curve image was processed by Mathematica 8.0 to get an energy distribution function.

Α,ί.2,5 ;sSsis^;5yss¾ in ows

Fig. A.1.2.5-1 show the PVC absorption (1-transmission) data in the [2.8, 5] um range from Yangzhou university. We then use two basic relations:

Emitter x Absorption(PVC)→ PVC s part

Emitter x Transmission(PVC)→ Left for water to absorb

{Emitter, PVC's part, Left for water to absorb} are overlaid in Fig. A.1.2.5-2. A numerical integration (performed using Mathematica 's NIntegrate[] function) shows that

PVC s part

— x 100% = 77.5476%

Emitter

Left for water to absorb

— x 100% = 22.4525%

Emitter The complementary relationship is very clear in Fig.A.1.2.5.

The continuous spectrum of 22.4525% "Left for water to absorb" part is then again integrated with the discrete HITRAN gas mixture absorption lines in [2.8, 5]μπι to get the actual portion of energy that the gas mixture could absorb. The integration is done twice, both forward and backward, and we pick the larger of the two as another fold of "safety measure" so that we always calculate higher temperature of gas than what would happen in reality. The "forward integration" is in fact done by

and the "backward integration" done by

The difference is whether we multiply an absorption line's absorption value to the energy fraction associated with the continuous fraction [λ[ί], λ[ί + 1]] or [λ[ί— 1],λ[ί]].

Note that power(A) is not the original IR emitter energy function, but is the lowest, "left for water" part shown in Fig.A.1.2.5-2.

The spectral lines between [2.8, 5.8]μπι contains 537583 out of the total 4172592 spectral lines over the range

[0.497512, 99.9989]um. The result was forward integration: 0.00043191 backward integration: 0.00043197

The two values are essentially the same, and with the large value, the ratio that of the IR energy that PVC absorbs to the energy that the gas mixture absorbs is

77.5%

8009.46

0.00043197 x 22.4%

So the PVC gets over 8000 times the energy than the internal air. The volumetric density between PVC and air is about

PPVC = 1-39 g/mL

1112

p_air = 28 _S/22.4 L and their specific heat (PVC 1.39 vs air »l) are essentially on the same scale. So the 8000 ratio means the energy allows us to heat PVC of around— — = 5.176 volumes of that of air while still maintaining the same temperature rise. The PVC surrounding the internal is only 0.5mm, and its volume is only a small fraction of the internal air, so up to this point it is almost intuitively clear (based on the quantitative results) that the PVC will always be maintained hotter than the internal air.

The above result was superbly well, but it is also specific because we have particularly chosen the IR emitter whose peak band matches that of PVC ' s peak absorption. In the following we provide a worst case estimation of the minimum energy ration required.

How large should the ratio absorption (PVC) / absorption (gas)be in order for the PVC drip chamber to be hotter than the gas? A very rough estimation can be done by assuming that the IR energy PVC absorbs is only used to heat a very small area where the IR beam is focused at, due to its poor thermal conductivity, and the IR energy gas absorbs is used to heat the gas in the entire gas chamber.

First we calculate the specific heat of the humid gas mixture: = 1.006//(_S. °C), cwater vapor ⁼ l-864//(#. °C)

The water mole fraction » 5%, so c_mix = 1.006 x 95% + 1.864 x 5% = 1.0489 J/ g. °C) As having been calculated previously,

28 g/mol

m_air ~— x 6.1575mL = 0.0076969 g = 7.7 mg

So for all air to raise AT = IK,

AQ_air = c. m. AT = 1.0489 x 7.7 x 10^"3 x 1 = 8.07653 x 10^"3/ whereas for PVC, if we heat specifically an area of 4cm² and use measured thickness of 0.5 mm, density 1.39g/mL and specific heat 1.5 J/(g. °C) as in Table. A. 12.7 of [Wilkes, Summers, Daniels, PVC Handbook], then for it to raise the same IK temperature,

AQpvcarea = c. (ρ. ί. Λ). ΔΤ = 1.5 x (1.39 x 0.05 x 4) x 1 = 417 x 10^"3/ AQPVC area ₌ 417 X 1(Γ³/ ₌

AQai_r 8.07653 x 10^{" 3}/ so the conclusion is that if we use radiation heating to heat only a 4cm² area, ignoring heat conduction within the PVC itself and convection to surrounding air (both external and internal, since the simplification is that the internal air is heated by radiation only), then a ratio of 52 would ensure that the heated PVC area to be constantly hotter than the internal air. This is of course a gross oversimplification, but give us a meaningful sense of the actual energy fraction ratio required for this application.

We also did the simulation in Solidworks by using 8mw, 16mW and 24mW energy input from the IR emitter. The modeling was different from direct contact heating:

1. For PVC, a "patch" of 4.56mm(W) ^x 4mm(H) was used as a volumetric heat source having _{g 0 0 6} of the 8mW, 16mW and 24mW energy respectively. The material was set to be the same as surrounding PVC. Please refer to Fig.A.1.2.8-1.

2. The remaining °— 801 -0—.46 fraction of the energ ^Dy^J the g oas mixture g oot was modeled as a 1mm diameter tube set as a surface heat source.

It should be understood that the above setting makes sense. The Patch was chose to be 4.56mm(W) ^x 4mm(H) because it is there are lens (IR specific optical material, like Germanium) and reflectors in the IR range capable of providing parallel beam from the emitter source, so we can create a relatively wide beam to heat a wider area of the drip chamber which would be free from splash/condensation droplets, and that would be the "view window" of our camera. On the other hand, we should not use a very thick tube to model the IR absorption of gas mixture because that would cause a significant change to the internal gas volume. Because of the 8009.46: 1 ratio of power absorption, the condition on the PVC wall dominates and the tube surfaces actually have little effect on the eventual overall temperature distribution.

Also note that in Fig. A.1.2.8-1, the we only used one such "patch" to model the front (first) incident surface of IR upon the drip chamber, but not the second. This is because the majority of the emitter power (77.5%) will be absorbed at the first drip surface, and ignoring the 0.00043197 gas mixture absorption of the remaining 22.4%, practically another 22.4% x 77.5% energy would be absorbed by the back surface. But whereas the front surface in such configurations is surrounded only by air, the second (back) surface might be surrounded by a solid (metal, plastic, etc.) supporter/fixture, and it will dissipate heat much faster than the front surface. In addition, we only need to prevent splash/condensation in the front surface because that is where they will block/obscure our view. Even if there are droplets in the other side of the drip chamber surface, they could be removed easily using background removal algorithms in image processing.

It is possible to create 16mW and 24mW emitter source by using multi-die LED technology. The [A novel LED module for the detection of H2S at 3.8μπι] paper was published in 2000, and clearly the technology available today would give us greater freedom in customizing the emitter characteristics.

Results Heated PVC area temperature (steady state), Time taken to reach steady state ambient air = 25 °C

8mW 309K = 36°C

16mW 320K = 47°C reached 45°C in 5.85 minutes

24mW 327K = 54°C

Also as in the case of direct contact heating, hot air remain in the upper area, and the lower part remains cool. The 47°C and 54°C temperature would be felt by our finger to be pain, and especially for the 54°C temperature, it would be difficult for any splash/condensation droplets to exist for an extended period of time in the vicinity of the heating center. d

We also conducted time-dependent simulation of the model (steady-state ignores the— term in the partial differential dt

d

equations whereas time-dependent analysis considers— term and calculates the time-dependent state evolution). For dt

the 16mW case it takes 5.85 minutes to reach 2°C below equilibrium temperature, and the actual convergence to steady-state temperature might take quite long. He who uses this invention should choose the appropriate emitter output power to achieve his desired temperature rise speed. And when it is desired to reach a certain temperature faster or slower the output power shall accordingly be adjusted.

In summary, we have shown with calculation and simulation that a pure radiation (without emission in the visible spectrum) source can be used to heat an area of the drip chamber to a temperature without heating internal gas mixture to higher than that temperature. The scope of our analysis techniques, apparatus and methods clearly covers all radiation sources, including the "equivalent IR heating bulb" mentioned above.

Note on the validity of PCT/IB2013/056090 computation:

Please note that for the above and subsequent calculations the Fresnel reflection loss, which later shown in increasing absorbance takes value 0.04258 in the visible range. Without knowing exact index of refraction in near and far infrared it is imprudent for us to use the index at visible range, hence we didn't account for varying n and Fresnel reflection loss in calculation. According to the calculated ratio above the dominant heating on PVC over gas would not be changed after accounting for reflection, and we have calculated and verified this for other cases in this application.

Wavelength Selection

Due to challenges in material, at mid and far IR semiconductor the efficiency of semiconductor/LED source is usually low enough (as the time of this application) to offset the high drip chamber absorption. Therefore, in addition to the much more efficient incandescent/resistive source discussed below, we propose using visible and near IR range sources, including LED sources, as the front side heater. For example, it is reported that OSRAM dragon IR emitter SFH 4232A can reach 39% luminous efficacy (depending on its definition of "luminous efficacy" outside visible spectrum; centroid wavelength 850 nm), which gives an example that near IR and visible spectrum (LED efficacy typically 15%-20%) sources (all, not limited to LED; for example sodium lamp might also have >20% efficacy) can also be used as front- side heater.

A.1.3 Heating by Illumination Source (Isieamksees . light)

Next we prove that it is also possible to heat the drip chamber effectively by an illumination source. As an illustrative example, we show how incandescent lamp can be used to prevent/remove splash and condensation droplets while simultaneously illuminate the scene of the falling drips for the camera.

Using an illumination source for heating differ from IR emitter heating primarily because of the different shape of the emitting spectrum. It is well-known that only 5% to 10% of its tungsten filament's emission spectrum energy is within the visible range, whereas all other parts are wasted as UV and infrared energy. Neither of PVC and gas mixture (water, C02) absorb UV or visible spectrum energy (please refer to Fig.A.1.2.1-6: maximum gas mixture's absorption in the visible <2> 10^"4 at around 0.72um, and for smaller wavelengths the absorption are all on the scale of 10^~5; for PVC, very thick PVC is also transparent so the absorption in the visible should be very weak, and we have assumed 95% transmission rate for our 0.5mm-average thickness samples in our calculation for wavelength smaller than 1 μπι.

An experiment is shown in Fig.A.1.3.1-1. In the left of the figure there is a drip chamber having many condensation splash and droplets inside, and close to it is a vertically fixed (by the fixture in Fig. A.1.3.1-2) small incandescent lamp seat, having a small 6V light bulb (Fig. A.1.3.1-3) mounted on the seat and a cylindrical wrapper rolled from a piece of hard card paper wrapping the bulb and is also in direct touch with the drip chamber. The cylindrical hard card wrapper serves as both a light and heat blocker and director, for

1) Without such a director, the light from the bulb (its filament can be regarded as a point source) will cause very bright and uneven reflections from the drip chamber, and this phenomenon is common for all types of point or concentrated light sources. The mechanism has been described in detail in our previous application [US13356632, Image Processing, Frequency Estimation, Mechanical Control and Illumination for an Automatic IV Monitoring and Controlling system], particularly in section 4.1 and 4.2, and illustrated in Fig.A.4.1-1 to Fig.A.4.1-3. The bright reflection will cause problems to the camera, making it difficult to find the position of the drip in the captured image (see [App US12804163, paragraph 0099]), and create problems to subsequent image processing due to lowered image quality. For example, if we need to measure the size of the drip typically we need to detect its edge, and edge detection algorithms typically work by calculating image pixel value's gradient. Bright reflection from the light source could change the gradient field and cause subsequent problems. The director/Mocker will block spherical light rays going out in non-axis directions (see Fig. A.4.1-3 of [App. US 13356632 Image Processing, Frequency Estimation, Mechanical Control and Illumination for an Automatic IV Monitoring and Controlling system]), and the rays eventually reaching the illuminate/heated surface are more close to being "parallel" than being "radiating", and this will significantly reduce the uneven shinny reflections.

2) Such a cylindrical wrapper also confines much of the energy inside its area. The incandescent bulb shown in Fig.A.1.3.1-3 is the simplest type of spherical bulb, having no light directing mechanism like a reflector (for example, "PAR" Parabolic aluminized reflector light or MRx(opening diameter) Multifaceted reflector light) or lens, so the radiation goes in all directions, and some basic geometric calculation will show that an overwhelming majority of its radiation goes to the directions that will never touch the drip chamber. A cylindrical straight director/Mocker forces the light to go along the director of its axis, and even for the majority of the light that undergoes multiple reflections on the director/Mocker's surface (this would be particularly if the interior surface of the director/b locker is made of polished metal, like aluminum or silver), it would eventually come out of the cylindrical volume at the exit, still delivering much of its energy to the object being illuminated/heated. In this way, the director/Mocker serves as an energy transmitting channel.

After we had turned on the lamp, it began to heat its surrounding both by infrared radiation and by convective heating of the surrounding air. In about two minutes' time, all splash and condensation droplets originally seen on the side of the chamber touching the card paper director/blocker disappears, as seen in Fig. A.1.3.1-1(2). The experimented continued until over 500mL of liquid dripping has finished and during the entire course the area under the direct illumination/heating of the 6V light bulb never showed any significant droplets (tiny splash droplets all disappeared very soon).

The experiment above is an experimental demonstration that we can resort to the heating power of illumination sources to prevent/eliminate splash and condensation droplets. However the characteristics of illumination sources vary so relying on the instance of merely one experiment is chancy. To firmly establish the method's validity we also need to calculate the energy absorptions of the PVC drip chamber and gas mixture quantitatively.

We obtained the following measurements of the bulb:

The incandescent light's total power is

UI = 6 x 0.45 = 2.7 Watt. Without considering the energy that is dissipated through the bulb's electrical contact with the seat, the remaining energy is all dissipated through radiation of the filament (the inside of the bulb is in general vacuum so heat could not be through air conduction). The bulb absorbs a portion of the energy primarily in the IR band and itself gets heated (100°C to 110°C), beginning to radiate IR energy as a secondary IR source. Besides, the bulb also heats the surrounding air be conduction and convection.

Our simulation in Solidworks Flow Simulation shows that in such a partially enclosed environment (the card paper channel in Fig.A.1.3.1-1(2)) without forced convection (like a fan) and with very poor thermal insulation (card paper only), the amount of heat transferred by air is small compared with radiation. Therefore we primarily analyze the two radiation sources:

1 ) Radiation from the tungsten wire

2) Radiation from the glass bulb

The normal working range of tungsten wire is between 2500K and 3500K, and because it is difficult to determine the actual wire temperature inside the bulb (even with thermal imaging devices, the reflection of glass complicated the situation and requires special tuning), in our following calculations we have used three values: 2500K, 3000K and 3500K calculated for cases.

What makes the analysis complicated is the existence of two "windows" (illustrated in Fig.A.1.3.3-1), the glass window, which absorbs and re-radiates tungsten filament radiation, and alters its spectral energy distribution, and the second PVC window, which absorbs both portions of both (glass-filtered filament radiation, and glass radiation), and leaves the remaining to the gas mixture. We have obtained the full spectrum (assumed 95% transmission for below ^m as in the second paragraph at section A. 1.3, and used FTIR to measure l-lOOum range) transmission data for the PVC drip chamber material, but for the bulb glass, due to

1) The spherical shape refracts IR rays so the measurement cannot be exact.

2) It was difficult to find a suitable fixture to hold the bulb glass piece (cracked) firmly and mount it in the FTIR spectrometer.

Therefore we didn't use FTIR measured data for glass transmission but have assumed three windows functions shown in Fig. A.1.3.3-2 to model its transmission behavior. Each of the three functions have a center and a defined width, and the their generating function is

The maximum transmission is 80% when λ = center, and when λ = center ± width/2, the transmission is actually 44.4%. The three transfer functions shown in Fig.A.1.3.3-2 are actually quite representative of a most common bulb glass materials (they are usually "practically opaque" in the infrared range and have a cut-off frequency between 2.5 and 4.5μπι).

In addition, because the temperature range of the light bulb was measured to be between 100°C and 110°C, we use three values, namely 100°C, 105°C and 110°C in our calculation.

Recall that in the discussion of the 3 8μπι IR emitter, the fraction of IR emitter's 22.4% is again integrated with HITRAN absorptions to yield the accurate gas mixture absorption figure. f 2(i +l)

power(/l)d/l

absorption[z] x— ^ι-

2·8/«") [ power(/l)d/l

J 2.8 and that power(A) = emitter(A)^■ [1— pvcTransmission(A)]

What we are doing here again for the gas mixture's absorption (percentage) is the sum of two terms: absorption(filament→ gas)

i(0.4975/mi) _oEmission_filament ( ) άλ for tungsten filament radiation's absorption; and absorption(bulb→ gas)

for glass bulb radiation's absorption (percentage). The gas cell was meant to be the same 1.4cm (inner diameter of a typical drip chamber) as in the 3 8um IR emitter experiment. For the transmission windows,

Transmission _glass (λ) is defined in above by the generating functions and we have indeed calculated for all the three modeled windows in Fig.A.1.3.3-2. The PVC transmission window Transmission_pvc(/¾,) uses directly the data from The Material Test Center of Nanjing University, and because the FTIR spectrometer measures only between 1 and ΙΟΟμπι, we define transmission for λ < Ιμπι to be 95%, and for λ > ΙΟΟμτη to be 0.1 based on the information in Fig. A.1.2.2-5. And just like we did in the calculations of the 3.8um IR emitter, here we also calculated both forward and backward integration for gas mixture's absorption lines and pick the maximum result as a "safety measure" so that we always calculate greater gas mixture than what it would be in reality.

The absorption (percentage) of PVC drip chamber is calculated as

and

In all of the above calculations, the Emission_filament(A) and Emission_glass (A) values are calculated using Planck's law:

Emission(A, T) = emissivity^■ and = 3.742 x 1Q⁸ W^■

C₂ = 1.439 x 10 m^■ K.

Also in all the calculation we used the complete 4172592 absorption lines of the gas mixture without any simplification (such as ignoring/skipping some lines with reasonably low values) to ensure the highest accuracy. Because of the large amount of data (175MB for the spectral lines, and numerical integration with each of them), the processing, cross- validation (by various repeated experiments generating redundant intermediate data for manual examination) and post- analyzing took us over five days to complete.

A.I.3.4 ScsisMs

The most important results are shown in Table. A. 1.3.4.

Table.A. 1.3.4 / Fig.A.l.3.4-5

In the column "glass→ PVC", we see that for all parameter combinations PVC consistently gets over 94% of the glass radiation power, this is because the temperature of the glass is much lower than that of the filament, and by Wien's displacement law its peak frequency is at

2897.8

= 7.66 am

378.15 ^ and from Fig. A.1.2.2-4(1) we see that the 0.5mm PVC samples are essentially opaque in that area.

In this column the value for fixed T_fllament and T_glass decreases as the glassWin becomes narrower and shrinks towards the left. This is because the more the window is to the left, the less IR energy from the filament it allows PVC to absorb.

View Factor and Solid Angle

Some might find that the values between 15% and 30% percent surprising large: How can such a thin layer of PVC absorbs so much of the filament's radiation? This is very counter-intuitive because we don't have a conceptual knowledge of what materials behave in the invisible IR band. In the visible spectrum, the PVC is highly transparent and for a thin layer of 0.5mm, it in fact difficult to perceive any appreciable light attenuation if we look through it against a light source. The reason that such an apparently "transparent" material absorbs as much radiation power from the filament is explained in Fig.A.1.3.4-1 for a combination of filament temperature 2500K and a glass transmission window centered at Ι.δδμπι β^ ΐΜβ a width (according to section A. 1.3.3 definition) of 2.5μπι:

First, in the upper most sub-figure we see that its peak radiation is at A = 1.1592μ?η according to Wien's displacement law. Its power distribution over several regions of interest is

So we see that in fact only a minor fraction of below μπι, and only 5.5% in the visible spectrum where we could perceive. So even if the material (PVC here) is completely transparent in the visible range, its weight in the overall absorption would so small as to be even negligible.

The glass transmission window and the PVC absorption windows in the [0, 10] range is overlaid in the second sub- figure. The PVC absorption under 1 μπι is low (because we assumed 95% transmission, but if we do measure real data which arguably should be smaller than 95% between [0.769, 1.0] it will have very little influence on the overall scheme), but between [1.0, 3.5] where the glass transmission falls from as high as 0.8 to as low as 44%, the PVC absorption is at least 0.2, and can reach as high as 0.6. The minimum absorption considering the cascading of two windows, is therefore 0.8_{glass transmission} x 0.2_PVc absorption = 16%. Over the [1, 3.5] band there are as much as 72% percent of the filament of radiation, so at least 72% x 16% = 11.52% is absorbed by the PVC as lower limit estimation.

The joint effect is glass transmission and PVC absorption is shown by their multiplication, and is plotted in the third sub-figure. The 4^th sub-figure is the multiplication of the 1^st and the 3^rd sub-figure, in which the shaded region represents the portion that the PVC could absorb. The accurate figure is 18.5642%.

2.7W x 18.5642% = 0.501 W

We therefore see such a significant absolute value of power absorption. Recall that for the 3 8μπι IR emitter, our Solidworks Flow Simulation model a patch of 4.56mm(W) ^x 4mm(H) size absorbing 24mW power could raise its surrounding temperature to over 327K, or 54°C, and is significant enough to evaporate any splash or condensation droplets. An absorption of 0.5W is over 20 times that value, and it would have such a potential as to even melt the entire PVC material.

Why the melting didn't happen in our real experiment (Fig.A.1.3.1-1)? This is all due to the small solid angle ω that the PVC drip chamber surface extends with respect to the view of the spherical light bulb (see geometry Fig.A.1.3.4-2). According to Table.A.[1.3.2-l] the card paper wrapper has length 18.43mm and its "mouth" in contact with the drip chamber is measured to be about 3mm away from the leftmost point of the bulb sphere. The radius r of the bulb sphere is 10.92/2 = 5.46mm, so the solid angle of the filament's radiation (ignoring its shape and treat it as a point source) that could reach the PVC drip chamber corresponds to a small "cap" with base radius 5.46mm on a sphere of radius 8.46mm. The solid angle can be calculated using basic calculus as:

2π x (l - Sin[ArcCos[5.46/8.46]]) = 1.48375 which is only

1.48375/(4π) = 11.8% of the total 4π solid angle of the sphere. So basically, only about 1/9 (0.0591M/) of the filament's radiation can reach the circular "mouth" of the card paper wrapper, the remaining are all directed to other directions:

1) Some blocked, absorbed and re-radiate by the card paper wrapper

2) Some transmitted through the card paper to the outside

And for the 0.0591M/ which reached the surface of the wrapper "mouth", yet another some radiation rays will not touch the drip chamber because the wrapper and the drip chamber cylinder are conforming seamlessly to each other; even for the rays that could touch the drip chamber surface, most of them will have an oblique incident angle so that only a certain percentage of them could be absorbed. When all these reductions are considered together, the overall heating effect would not be quite different from the ideal straight irradiating 3.8μηι IR emitter source.

Papers generally have emissivity above 0.9, and effectively behave as a third radiation source. However, its structure is highly porous (have lots of small cavities) and it is not easy to determine how much radiation it can absorb from the filament and bulb glass, which are essential in calculating it re-radiation (some will reach the drip chamber). However, its characteristics are unimportant since it is only a make-shift implement used for this particular experiment, and in real implementation people will definitely use other material with more predictable characteristics (for the reliability required by a medical device). What is important is the finding that strong radiation absorption can be expected from ordinary types of PVC drip chamber material.

The significance of this finding lies in the fact that filament radiation is the primary power source of an incandescent light. The absorption and re-radiation of the bulb glass in only secondary, and constitute only a fraction of the former. The power ratings of incandescent can be as low as to at least 0.2W for miniature ophthalmic lamps, so even without the diligence of making incandescent lamps for our special needs but only use these existing miniature lamps, we can still always get the desired heating power. For example, assuming 90% of the 0.2W is at first radiated from the filament, and assume a filament temperature of 2500K (fairly low and will result in extended filament life time), and use the same 1.55μπι-οβηΐ6Γ, 2^m-width glass window as in our previous calculation, so that 18.5642% out of the total 0.2W x 0% = 0.18M/ radiation, i.e.0.0334M/, will be absorbed by the PVC drip chamber, which is already greater than the 24mW value which resulted in over 50°C drip chamber are temperature used in our 3.8μπι IR emitter experiment. The waste of power in other directions can be solved by using directional light, which is achieved by using reflectors, for example parabolic reflectors (PAR, parabolic aluminized reflector light, etc.), ellipsoidal reflectors (MRx, multifaceted reflector, etc.), general aspheric/spherical reflectors or even a back/lateral side mirror. For parallel or quasi-parallel light beams shaped by these reflectors, it is reasonable that over 80% of their energy will be absorbed by the PVC drip chamber.

Halogen lamps: Halogen lamp is a type of incandescent lamp and differs in that the halogen gas contained the bulb re- deposits evaporated tungsten back to the filament. It is clear that all our analysis techniques apply to them. It is possible that some (in general slight) modification to figures and conclusions might needed, but there are not essential difference

Coating

Advantage

Our calculation has revealed that the use of incandescent can solve the two problems simultaneously. Approximately 5% of its energy will be in the visible band an can be used to illuminate the source, yet this is only true if we restrict ourselves to visible band cameras. In our video-monitoring based infusion device we are taking images in an enclosed environment, so it is entirely permissible to use digital camera whose sensor are tuned in a particular IR range (not the UV range because UV light cause adverse effects to certain drugs) where the absorption is not strong, it we deem that the 5% visible range illumination is not enough. But after all these are not problems, for:

1) We can choose or build digital cameras with high sensitivity. Because we are shooting in an enclosed environment where all parameters are known, customization could be very easy.

2) The brightness of images output by the digital is not determined by illumination strength alone, but also "aperture size" and exposure time. [Please refer to Hecht, Eugene - Optics 4ed section 5.3 for "aperture size"] . If we increase aperture size and exposure time, even with dim sources we could get bright image. Also please refer to our App US12804163 IV Monitoring by Video and Image Processing, paragraph [88, 89, 90, 107], it is advantageous to focus on the much slower "forming" process of the drips when drips grows from smaller to larger and eventually falls, so increased exposure time won't be a problem when shooting this slow process. In addition, we have another controlling parameter, the Gain factor, which is essentially a signal amplifier/multiplier, and if the SNR (signal to noise ratio) is high enough, increasing Gain will give us very good quality image with increased brightness (intensity).

Life-time

The figures from US department of energy generally cite incandescent bulb life time between 750 and 2000 hours (for example [http://appsl .eere.energy.gov/buildings/publications/pdfs/ssl/lifetime_wMte_leds_augl6_rl.pdf]), and many lamps from GE/Philips list life time of between 4000 and 8000 hour. The life time of incandescent lamps are influenced strongly by voltage, and reducing voltage by 5% from nominal voltage can increase the life time by 100%. This relationship is described in [Coaton & Marsden - Light Source and Illumination, Ch8 "Incandescent light", section 8.4], and is expressed by

Life o V^{~L (}-^V)

Where L(V) is a exponent ranging between 11.8 and 14.5 for voltage decrease no more than 10%. For L (V) = 11.8, 12.7, 13.6 and 14.5 the life increase when V changes from 100% to 90% of luminal is plotted in Fig.A.l .3.4-3. From the plot we see that if we decrease voltage by 5% or more, it is not difficult to extend the life time to 2, 3 or even 4 times.

Multiple spare lamps

We also recommend assembling multiple incandescent lamps at a single illumination location. Miniature incandescent lamps can be made fairly small (thinner than ¼ of a little finger), so we could assemble 3 or 4 into as a single "bundle", and use only one or two at a time. The detection of which fails can be achieved with simple circuit and the software will re-enable others.

With all these measures taken, we could expect the life time of a single bulb to be over 4000 hours, and for 3 or 4 of them significant over 10000 hours, which is over 3 years based on a 10-hour day usage. Replacing a component every 3 years during maintenance is certainly acceptable.

Non-obviousness

The use of video and image processing techniques to measure and control drip flow rate is a new area, and there are a very limited number of prior arts. The earliest we are aware of was from China patent application 200710168672.3 from Song, Enmin, titled "A medical infusion speed monitoring and controlling system" which mentioned the use of LED lights but not incandescent light; another prior art was App US12791885 "Intravenous Drip Monitoring Method And Related Intravenous Drip Monitoring System" which did not mention light source at all; the latest prior art is App US 13631422 by Michael G. Lowery, which mentioned LED multiple of time but didn't refer to incandescent light. Aside from these patent applications, to our knowledge there are also no available devices on the market using video/image processing techniques to monitor and control IV dripping process (and hence no use of incandescent light on them).

There are multiple reasons supporting the use LED as illumination sources:

1) Efficiency: LED 's are much more power efficient than incandescent. Whereas only 5% and 10% of incandescent light's power is used to produce in the visible range, the efficiency can reach over 30%. An 0.2W incandescent light as we have analyzed before would require as much as 54mA of current from a 3.7V lithium battery, yet for a typical portable medical device such as the one we show in App.

US13903924, the entire board power, including the processor, DDR2 memory chip, camera and LCD display, is only 70mA. Modern semiconductors are very power efficient, and it is always the goal of designers to make the current consumption as small as possible. Even a miniature incandescent lamp of 0.2W increase the power consumption by 77% (124mA/70mA), meaning that the battery life would be reduced to only 56% of without such (arguably the smallest) incandescent light source and would almost certainly be optioned by designers.

2) Life time: LED typically have life time between 35,000 and 50,000 hours, 17.5 times higher than an 2000h-life incandescent, and still many times higher after we apply techniques such as reducing incandescent light's voltage. 3) Chromatic aberration: the refractive index _; differs with wavelength (frequency), so a single ray consisting of light component of different wavelength will be refracted differently after going through the lens system, this is illustrated (exaggerated) in Fig.A.1.3.4-4. The lens design method to counter this effect is by using a convex-concave (usually a convex crown glass and concave flint glass) to cancel the chromatic aberration of each other. However, the emitting spectrum of single color LED is concentrated near a single spectrum and this could lead to considerable simplification in lens design, and aid to the improved accuracy of droplet size measurement.

And these advantages are widely known to electric and mechanical designers, and this helps to explain why the above prior arts unanimously suggested LED as their illumination source.

On the other hand, the final the conclusion that incandescent light can be used for heating PVC chamber yet without heating the internal gas mixture to a higher temperature, was reached after a serious of rigorous reasoning and diligent calculations, many of which are again our initial intuitions, and are out of the knowledge domain of experts in any specific technical fields.

The key factor enabling its use is a transparent material(PVC)'s non-transparent property in the IR range, and a particular material's IR absorption property is usually only known to experts in material science or those specializing in plastic researches, and is distant from the general field of video and image processing. PVC materials contain different levels of plasticizers among other additives, and the exact absorption spectrum of a particular type of PVC (rigid or flexible, which type of plasticizer) can only be measured using spectroscopy devices (requires knowledge of ATR (Attenuated total reflectance) and transmission spectroscopy), and we have spent a lot of effort in obtaining these data (from three universities, the near, far and long full spectrum IR absorption data).

Yet the absorption property of PVC alone is not sufficient to establish the technology's validity because we must consider how much the humid gas radiation the humid gas mixture will absorb, and to what degree its temperature will rise. This involves the knowledge of thermodynamics (blackbody radiation, phase change, condensation nucleation), the use of gas spectroscopy data (the generation of HITRAN (High Resolution Transmission) data actually involves quantum- mechanical calculations), heat transfer in gas and solid (the inventor is familiar with these differential equations), fluid dynamics, and the use of a CFD (computational fluid dynamics) simulation software. Our analyzing steps comprises first constructing PVC (using FTIR measured data) and glass transmission windows in Mathematica, then multiplying the windows and doing both continuous and discrete (both forward and backward) integration for the 4172592 spectral lines, and repeat the process for a total of 27 parameter combinations ([3 filament temperatures] ^x [3 bulb glass temperatures] ^x[3 glass transmission windows]). The intermediate calculation involves number out of the range of single-precision floating point numbers (as small as 4.8248 x 10^~45) and the multitude of data (175MB gas absorption data) defied our initial attempts of doing the calculation using spreadsheet programs.

The data obtained from the front end Mathematica processing, particularly the ratio of radiation absorption between drip chamber surface and gas mixture, was then supplied to the CFD software (Solidworks Flow Simulation), and in general we do both time-dependent and steady-state analysis. The use of CFD simulation is necessary because the complexity of the problem defies the direct use of classical analytic solutions, and without such simulation our conclusions could not be established firmly. We have taken approximately one month's time to study the relevant knowledge, and have consulted experts including researches in metrology (their research involves cloud condensation which is relevant to our problem). We believe these efforts are necessary, because patent law provides that one is only entitled to an invention if he discloses it in "such full, clear, concise, and exact terms as to enable any person skilled in the art or science to which the invention or discovery appertains, or with which it is most nearly connected, to make and use the same". As a consequence of this requirement, the rigorous calculation and proof we presented above are certainly required.

We continue our previous analysis of Table. A. 1.3.4 results:

This also decreases with glassWin[] transmission function shrinking to the left. This is also expected because the more the glassWin[] is to the left, the more filament IR radiation it prevents from escaping the bulb, and the gas mixture will certain get less (assuming the same bulb glass temperature).

Its value is a constant for fixed T_fllament and T_glass regardless of glassWin shape. This is also expected because the radiation from the glass surface does not have to go through the glass itself, but only needs to pass the PVC window.

The most important ratio from Table. A. [1.3.41 is ^f'^lament→pvc _rati₀ because the filament radiation is always the

r filament→Gas ^J

primary radiation source. Its minimum value is 287.11 (3^rd row), and we should use this value as the "worst case analysis" input to the simulation software to test gas mixture will be heated (by radiation) to hotter than the drip chamber.

The simulation used the same model as for the 3.8μπι IR emitter. This should be understandable because when constructing the real device we could use lens and reflectors to direct radiation to a small area on the drip chamber. The use of thin tube to model gas absorption is also explained in section A. 1.2.8.

The simulation assumed that the PVC drip chamber receives a radiation power of 0.03M/, and used the "worst case" 287.21 ^{fllament→pvc} _t0 calculate the glass absorption of 0.00012186W. The 0.03M/ value is close to filament→ PVC filament→Gas ° ^r

value 0.185642 times an 0.2M/ input. The reason that we are not using the values of in our 6V bulb experiment is because it served only as an experimental proof of the heating method, and the calculations later involved has already gained generality (by assuming different filament temperature, bulb glass temperature, glass transmission function, etc.), so we could use the figures obtained in Table. A. [1.3.4] to simulate the what will happen on a real product.

The simulation results are shown in 1.3.4-6. The temperature reaches 348Λ^" (75°C) at the center of the absorption patch, and remains high in the surrounding area. However the gas mixture's temperature is almost unaffected especially in the

filament→PVC lower, and hotter air remains in the upper region. This simulation has confirmed that our "worst case filament→Gas could effectively heat the drip chamber area to prevent and remove splash/condensation droplets, but will not heat the internal air to higher temperature than the chamber's inner surface.

As for the reason why we choose 0.03M/ as the input, but not 0.2 x 0.185642 ~ 0.037M/, we actually have conducted the simulation, and found that power will increase the small patch area of drip chamber to over 80°C, yet still have only an negligible on the temperature of the air. Though 80°C is still far below PVC's melting temperature, we feel that it is still more than enough. In building the device if we feel that the temperature might get too high for the heating area, we can increase the beam angle, so that heat energy will more widely distributed, resulting in heating a larger area more uniformly to a lower temperature than with a smaller beam angle.

Bulb glass radiation

We ignore the bulb radiation because it will in general be very small for directional bulbs in real devices. Radiation depends on both bulb temperature and surface area, and for low power (such as a 0.2 W miniature ophthalmic lamp) incandescent lights its temperature cannot reach above 100°C as for our 6V bulbs, and 60°C will be a good estimate. Still assuming glass emissivity of 0.92, then the emissivity power is eaT⁴ = 0.92 x 5.67 x 10^"8 x 333.15⁴ = 642.58W/m² but the glass bulb area is only on the order of a few square millimeters. A 25 mm² area will radiate 0.0016M/ radiation, and the surface radiation is diffuse (in all redirections), so even if we put the bulb in the close vicinity of the drip chamber, the radiation power reaching the device would still be smaller than 0.001M/, and from our previous calculations we know such power is not enough to heat even a small an area on the drip chamber to our desired temperature. Therefore we do not include it in our simulation model.

Because the PVC material of the drip chamber is a smooth and highly reflective material, illumination source (or incandescent light used simultaneously for heating) could also undesirably create undesirable reflection spots in the view of the camera. This is illustrated in Fig.A.1.3.5-1, and various solutions to the problem have been discussed in great details in our App [US13356632 - Image Processing, Frequency Estimation, Mechanical Control and

Illumination for an Automatic IV Monitoring and Controlling system} in chapter "Illumination".

In consistency the principle and uses of methods of US13356632, we disclose here another way of giving optimal illumination to the drip chamber.

The first of these methods is shown in Fig. A.1.3.5-3, where light(s) illuminating the drip chamber is enclosed in a structure having concave top-view profile for the surface covering (in front) the lights. The "concave" refers to both smooth (left of Fig. A.1.3.5-3) and convex (right of Fig.A.1.3.5-3) cases. The defining characteristics is The lights are placed (for example inside a cavity) in a structure having concave top-view cross-sectional shape near or touching the drip chamber. The concave inward bending surface serves to block the refraction from the drip chamber surface so the camera gets a clearer image.

The concave inward surface might or might not touching the surface, and if the structure is made of rigid material, for avoiding altering the shape of drip chamber (the lens formed by PVC drip chamber, see section A. 2.2) it may have fringes of flexible material (rubber, cloth, flexible plastic, etc.) to touch the drip chamber for fully covering the escaping reflection rays. Even it is not in direct contact with the drip chamber, the purpose of for blocking reflection rays should be evident from its structure and positional closeness to the drip chamber.

It is preferably used on the two "lateral" sides, as opposed to the camera which is facing the drip chamber because light will they symmetrically illuminates droplets edges on the two lateral sides, and these edges are the most important features for image processing algorithm's volume measurement. However, a single such structure can also be used for because of the symmetry of the droplet the other shape of the weakly illuminated side can be inferred from the shape of the strongly illuminated side (facing light).

Ray blocker on glass

It is known that for any directional light (LED, incandescent, etc.) the light intensity is strongest in the axial direction, and indeed the half beam angle is defined as where the intensity decreases to half of the axial direction strength. From this we know that the main contributor to brighter spots is the central parts of the ray.

Therefore we could partially block the central rays, and keeping other rays unblocked, to achieve relatively "softer" illumination rather than the "hard" effects of the central beam. In its simplest form, this is by placing a light strength reducing material, like a "patch", in center of the directional light's glass (or plastic), as shown in Fig.A.1.3.5-4(1 & 2). As long as the central beam strength is reduced, either partially or completely, the "softening" effect is achieved. Therefore this "patch" can simply be made of light absorbing material to absorb some light according to Lambert's law.

In an alternative form, the "patch" allows the light to pass through, but diffuse it in random directions for "softening". For example, sandblasted frosted glass can be used to achieve this, and the abrasion should only be applied to central region of the glass rather than the whole so that the peripheral rays are still directional. This is shown in Fig.A.1.3.5-4(1) where we intentionally texture the patch surface to look "diffuse".

In a preferred form , the "patch" is made of a reflective coating (Fig. A.1.3.5-4(2) where we draw the patch dark) so light (when we speak of light for illuminating we refer to light in the visible spectrum of the camera) is reflected back. In the case of incandescent or halogen light, this reflected light could serve to increase the power of the filament. Its distinction from prior art reflective coating is that the visible spectrum light (according to camera), rather than the invisible, is reflected, for the distinct purpose of reducing/blocking central beam strength. A, 2. Optical (Lens) Correction

All the previous heating and illuminating apparatus are essentially "paving the way" for the most critical step: optical measurement. We will prove in the following that indeed very high volume measurement accuracy can be achieved with cheap PVC IV (as low as $0.16 each, as opposed to $3-4 special resilient IV sets for infusion pumps) sets, therefore our device shows a strong competitive advantage with traditional mechanical IV pumps.

The description below involves knowledge in optics and lens design. The standard texts in this field are [Hecht, Eugene - Optics], [Smith, Warren - Modern Optical Engineering], [Kingslake - Lens Design Fundamentals]. We use Zemax for lens design and characteristics analysis, its use is documented in great detail in its User's Guide and online knowledge base of RadiantZemax.com. [Geary, Joseph M - Introduction to Lens Design with ZEMAX] is also a useful book on this software. We have used Zemax Programming Language (ZPL, see chapter 22 of its 2009 version user manual) to automate the generation of simulation images for different lens parameters, and use Mathematica to further analyze results.

A.2.J Blocking effect

Can we accurately measure the droplet's volume from 2D image? The affirmative answer is given by Fig. A.2.1-1. In the middle is an illustrative lens, and on the left are the side view of two concentric droplets. For the larger droplet, the ray from its topmost point might not be able to reach the center of the lens because it is blocked be some of its parts closer to the lens, even they are nearer to the optical axis, and the ray corresponding to the widest angle is the tangential ray (4^th ray, top most in the figure).

But in real system we always have non-zero entrance aperture diameter, and because droplet's volume is small (diameter almost < 5mm, determined by surface tension), the aperture can allow rays from the topmost point to pass without being blocked by nearer points.

In addition, the scale relationship in Fig. A.2.1-1 is highly exaggerated. In real systems the droplet's diameter is only a small fraction of the object's distance to the front plane of the optical system, and from the geometric relationship in Fig. A.2.1-1 it would be clear the blocking effect will become negligible as the lens becomes "far enough" from the droplet. Assuming the droplet has spherical shape and a 2mm radius,

2 x d² - 2²

h = 3

and this relationship is plotted in Fig. A.2.1-2. When the effect of wider entrance aperture is considered, there is practically no error introduced due to the spherical self -blocking effects. In addition, even in the worst case this very weak effect does exist, the relationship is simple and can be easily calculated and corrected. ¾| PVC JLens

What makes this problem distinct from ordinary 3D object size measurement is the effect of PVC drip chamber. The refractive index _; of PVC is 1.52, (refractive index is determined by atomic properties of matter, see [Hecht, Eugene - Optics 4ed, section 3.5 "Light in Bulk Matter"]), and even with the existence of plasticizer the change is also very small (can rarely reach 0.005). We have done extensive simulation in Zemax and confirmed that the effect of slight change (up to 0.02) of refractive index on the measurement accuracy is so small that it is not even detectable, and for all practical purpose we can safely use a constant value of 1.52. Of course, we could also use an "IV set library" to get data provided by IV set manufacturer, this will be discussed later. This 1.52 refractive index is higher than that of water (1.33). In addition, the thickness for our samples varies between 0.5mm and 1mm, whereas the inner radius is usually between 5 and 8mm. Although it is tempting to treat the drip chamber as a thin lens due to its 0.5mm diameter, it is actually not. For the most typical 0.5mm-thick, 7mm-inner radius drip chamber, the thickness is more than 7% of the radius, which is already out of the realm of "thin lens" [Hecht, 5.2.3 Thin Lenses] and the well-known thin lens equation does not apply. In addition because it is actually a cylindrical shape rather than axial symmetric, nor does thick-lens equations [Hecht section 6.1] apply. Moreover, the radius of the droplet is usually larger than 1mm, and when comparing with 7mm inner diameter it is also clearly out of the paraxial region (Hecht, 5.43). Although we initially derived some relationships by assuming thin or thick lens property, later quantitative analysis shows that none of them applies consistently across parameter combinations.

The drip chamber actually forms a cylindrical lens. See Fig.A.2.2-1, only the smallest of its numerous cross section planes is circular, all others are ellipses. The can be visualized by imaging a cross section plane's projection downward to the circular plane, and one of its coordinate is scaled and then both projected coordinates satisfy circle equations:

' = x^■ Cos[0], y' = y

Zemax software didn't come with cylindrical surfaces, so we actually have to create user-defined surface ourselves. This is done by modifying example C source files supplied with Zemax and change its surface drawing and ray tracing code, then compile to a new DLL (dynamic link library), and let Zemax load the file. Cylindrical lens alone with the remaining of the optical system is shown in 3D drawing in Fig.A.2.2-2(l), and Fig.A.2.2-2(2) shown an magnified view of the bending property of the lens. Fig. A.2.2-2(3) shows the a magnified view of the ray tracing through our cylindrical lens.

A.2 Optical LMS System

We use real lens in our study as opposed to "paraxial" ideal lens which Zemax also provides because paraxial ray- tracing simplification does not reflect real system behavior. The lens was designed for an Omnivision camera with 3 μπι pixel size, and the magnification ratio of the lens was -0.3 (inverted), so a 3mm diameter droplet will appear as approximately 900um and occupies 300 pixels, and this resolution is high enough for an accurate measurement. All these figures are of course with no intention for limitation.

The lens (Fig.A.2.3-1) was modified from a classical Cooke triplet by changing last two surfaces to aspheric. All the lens design data are provided in Fig.A.2.3-3, and the prescription data is shown in Fig.A.2.3-2. This lens has excellent MTF (Fig.A.2.3-4, modulation transfer function, how many fine lines in 1mm width can the lens resolve; over 120), spot diagram (Fig.A.2.3-5, how concentrated ray bundles are around the chief ray; <3μπι deviation, 1 pixel), ray fan (Fig.A.2.3-6, indication of aberration) and grid distortion (Fig.A.2.3-7, distortion from rectilinear projection; 0.6252%) performance.

A.2.4 Image Sim aiatton

Blocking effects: From prescription data in Fig.A.2.3-2 it is seen that entrance pupil (the image of the aperture, see [Hecht, section 5.3]) position is 33mm far from the droplet (image plane) and the entrance pupil diameter is 3mm, so from our discussion in "blocking effects" this effect can be ignored (and corrected if necessary). We will focus on analyzing how the system (PVC lens + optical lens) will image an object to the camera sensor (image plane).

First we construct a "source image" as shown Fig. A.2.4-1. The image was created in Zemax .IMA format, essentially a dotted array, and is documented in Zemax manual. We draw the concentric circles because this enables us to simultaneously study the optical system's distortion (due to varying PVC drip chamber lens parameters).

The core of Zemax's Image Simulation is the computing of Point Spread Functions (PSFs, similar to the impulse response of linear systems) for a grid of points, during which it considers diffraction, aberrations, distortion, relative illumination, image orientation, and even polarization. It will then place the source image at the object plane and convolve the PSFs with the oversampled source image to yield the final image.

A note on photorealistic lens simulation

There are other software capable of doing the same simulation. Code V is also widely used lens design software and gives reliable results. In addition, recent releases of PBRT (Physically Based Rending, from Matt Pharr & Greg Humphreys) also support tracing rays through real lens system. There is no essential difference if another software is used to do the same analysis.

We vary three parameters (see Fig.A.2.4-2):

1) The distance (D) of the droplet's mid plane from the inner surface of the drip chamber

2) PVC drip chamber's inner radius (R)

3) PVC drip chamber's thickness (T) The outside radius (highlighted in yellow) is the sum of R and T. The other highlighted term is the distance of the outside surface of the drip chamber to the first surface of the lens system, and it varies automatically with drip chamber inner radius and thickness.

We used Zemax Programming language to iterate through a range of these parameters:

1) For D, from 6.0 to 8.2mm

2) For R, from 5.5 to 7.6mm

3) For T, from 0.4 to 1.1mm

All these are realistic values. 7mm is a very typical R value and 0.5mm is common for its thickness. For UV-resistant IV sets T can be as thickness as 1mm. D varies between 6 and 8.2 because as PVC material is flexible, and the drip mouth's position can change under forces (Fig.A.2.4-3), or might be oblique due to manufacturing errors.

The results for two simulated images with parameters far apart ({D 8.1, R 5.5, Tl.l} vs {D 6.0, R 5.5, T 0.4}) are compared in Fig.A.2.4-4, and the lowest sub-figure is their difference. The difference in every concentric ring is significant. The horizontal width of the second image is 7.5% larger than the first, and its vertical height is 5.1% higher than the first. Estimating using ^nr³ formula, there is some over 15% measured difference between them.

The difference for images between continuously varying parameters is not as significant as for the extreme (far apart) example in Fig.A.2.4-4. We use Mathematica to further analyze Zemax's output images first compute its volume of each image using the fundamental slice-by-slice integration:

and only did that for the outmost rings. The results are shown between Fig. A.2.4-5 to Fig. A.2.4-7.

Analysis

Distance D :The most conspicuous information from Fig. A.2.4-5 is that the volume decreases with the droplet's mid- plane's distance to the chamber surface (due to bending, etc. like in Fig.A.2.4-3). The upper and lower sub-figure are plotted for the same sequence of data and the upper is used to show the changing of the D, R and T parameters. When D increases, V decreases from above 1.65 10⁷ pixel³ down to below 1.35 10⁷ pixel³, an over 22% difference. This is consistent with the rules of the lens system (both the real lens and the PVC lens) generating inverted real images: the larger the object distance, the smaller the image.

Radius R: with each fixed D, V increases with R. This is not as straightforward as D. We give an interpretation with thick lens formula (Hecht, section 6.1): = 1.52

1 _{i i}v ^{1 1} (« - 1) 0.5

(n - 1)( +— )

-r -(r + 0.5) n - r(r + 0.5)

n _en , -0.5 0.52 - 0.5 ,

0.52 ( + )

r (r + 0.5) 1.52 r (r + 0.5)

-0.17101

r (r + 0.5) and is plotted in Fig.A.2.5-1. Recall that cylindrical lens surface actually consists of multiple elliptic surfaces (Fig.A.2.2-1), and the radius of curvature for ellipse at the minor axis (b) is a²/b. Increasing R increases a hence a²/b, hence r (to differentiate from R) in the above formula. As the negative decreases in magnitude, hence the magnified, erect, virtual image of the droplet becomes closer and closer to the right (the real droplet), hence for the real optical lens we have a close object (the virtual image of the droplet formed by the PVC lens), hence the image has the tendency to grow larger.

The above interpretation is not complete. As the virtual image formed by the PVC lens gets closer to the real lens, it also decreases in size. There are numerous such elliptical surfaces and there is no simple way to add their effects expect by doing ray -tracing (tens of thousands, and can be millions), which is more accurate than reasoning with analytical formulas (which themselves are simplifications under certain restrictions). The un-decidability of the above analytical reasoning is in fact because the formula didn't include enough information for the particular problem which is an unavoidable limitation of the analytical branch of mathematics/physics (as opposed to the numerical/ computational branch). We cannot rely on a simple first order approximation in such a complicated situation involving the interplay of competing factors. As another corroborative example, we know that lens optical aberrations (spherical, coma, astigmatism, field curvature, distortion) can be analyzed (calculated) with at least three-order approximations. We there maintain that Zemax results are the more definitive ones that should be relied on.

Thickness T: Its influences are complex. It is more clearly shown in Fig.A.2.4-6 : for smaller D's (droplet closer to optical lens), increasing T increases V for both small and large R's; as D gets larger, the trend is that increasing T gradually changes to decreasing V for smaller R's, and for larger R's the change happens slower. This is most clearly seen in Fig. A.2.4-7, the rows from up to down corresponds to increasing D (from 6.3 to 8.1), and the columns corresponds to increasing R, each cell plots V change for increasing T's. The paradigm shift (increasing with T to decreasing with T) from top to bottom rows are clear in this Table. A.

Principal influencer

Fig.A.2.4-5 to Fig.A.2.4-7 visualizes the same experiment data set in different ways. Important figures are listed in the row and column headings for Fig.A.2.4-7: Max(V) -Min (V)

1) Row headier show for fixed D, X 100% for all combinations of R and T. For

Min (V)

value is around 5% for most D's, and reaches over 6.7% for D = 7.8 and 8.1.

Min (V) - V{T = 6.9)

2) Column headers show x l00% and

V(T = 6.9)

Max(V) - V(T = 6.9)

X 100% (T = 7 or 6.9 results are very close, and 6.9 is chosen because

V(T = 6.9)

the iteration increment for T was 0.2, so it skipped 7.0) for fixed R and different D, T combinations. The results are over 9% for most cases, and reaches as low as 8.4%, and as high as 11%.

From these values we see that D, the distance from the mid-plane of the droplet to the first surface of the drip chamber, has the strongest influence in V. This is more clearly in the lower sub-figure Fig. A.2.4-5, where the major changes are all due to D; and in the surface plots of Fig.A.2.4-6, where we see for any fixed R and T coordinate position in the R-T plane, V changes from around than 1.6^X10⁷ down to 1.3 ^x10⁷, more than 20%.

Therefore we now have the comprehensive information of D, R and T's influence on V:

1) D is the principal influencer of V (droplet volume) measurement. V decreases with increasing D.

2) For fixed D (and weakly T, with less certainly), V increase with R.

3) V-T relationship changes from V increases with T to decrease with T as D gets larger, and the change happens slower for larger R's than smaller R's.

Our inventions follow logically from the conclusions above.

For D

Control:

Shown in Fig.A.2.6-l(l to 3) is a gripper whose purpose is primarily for fixing D relative the optical lens system. Its defining characteristics is that

1) It consisting of two essentially symmetric parts (not necessary in shape, but in position to the ideal mid- plane of the droplet as illustrated previously) with respect the optical lens system's intended object plane (droplet's mid plane). The symmetry is for ensuring that the center of the object it wraps is coincident with the optical system's intended object plane.

2) It wrappers BOTH the upper and lower relatively harder and thinner ends of the drip chamber (as opposed to the softer middle part, because wrapping the middle softer part firmly will effectively change the PVC lens's diameter). This is for ensuring the drip chamber stands upright, not oblique. 3) The upper and lower part are rigidly connected belong to a same part (plastic, metal, etc.) so that they always open and close to the same angle. This is in fact required by 2) for without such synchronized same-angle rotation the drip chamber will be oblique.

Many drip chambers are assembled from plastic parts with different diameters, and the most important element in determining D value is the small "drip mouth" where the droplets comes out, and the outside structure of the drip mouth might not necessarily have the same diameter with the other "sections" (see Fig.A.2.6-l(3) annotations). Therefore optionally and preferably, the upper part of the symmetric gripper might be composed of part having different diameters to be able wrap the "drip mouth" structure closely. This is also illustrated in Fig.A.2.6-1(3).

Measure D & R: Secondary camera

D can also be measured, and this requires a second camera. In Fig. A.2.6-2(l to 2), a primary camera is for capturing images of the droplets for V computation, where as a secondary camera on the lateral side monitors the position of the drip mouth or droplet in the vertical symmetric plane of the primary camera. In this way we can obtain D even with the need for the firm gripper above, and even for heavy bent, or poorly manufactured IV sets.

Both cameras has the ability to measure R. The "blocking effect" (see section A. 2.1) can be corrected using the triangular relationship. The T can be measured using a combination of IR emitter and receiver by Lambert's law and is discussed below. Of course, the role of primary and secondary camera can interchange.

Adapt

The primary camera might also contain a linear actuator (motor, etc.) for changing its position (see both Fig. A.2.6-2(1 & 2)) relatively to the object plane to control D range for more accurate measurement. Auto-focusing micro-motor become increasingly popular in recent years on cameras, and such motors for example can be used for this purpose. However, the purpose of auto-focusing is only for gaining a clear (sharp) image according to some criteria, and is distinctly different from our purpose (and principle).

T (thickness) measurement

Recall that Fig. A.1.2.2-2 contain FTIR transmission spectrum for 1, 2, and 3 layers, and as we have discussed there, the different groups of spectral lines are clearly distinguishable. For single layer PVC the curves from different sample (different manufacturer) matches pretty well. For more layers the discrepancy is due to factors such as the technician didn't press the overlaid layers flat enough, and various artifacts. In Fig.A.1.2.2-2 there are a total of 14 transmission spectrums for 5 different samples.

The distinct shape of the transmission spectrum suggests that they can be used for T measurement. Because R can be solved when from the ratio between receiver's strength to the emitter strength. The drip chamber surface's reflection effect is certainly predictable and could be calculated. With this approach, R could be measured to a reliably high degree. The wavelength of the IR emitter can be chosen in any convenient range where two layers of PVC does not completely absorb the energy, it is clear from Fig. A.1.2.2-2 there are a wide range of choices. This measurement is method is illustrated in Fig.A.2.6-3.

Therefore, by introducing a lateral camera and an IR emitter-receiver pair, we are now able to obtain D, R and T parameter which are central to droplet 's volume measurement.

User input D, R, T and Refractive Index

Although the above methods have presumably high reliably, when manufacturer specifications are available we still prefer them as to re-do the calculation. Therefore a last method for obtaining D, R and T is either:

1 The IV sets containing identifiable marks, such as a special color means a particular {D, R, T}

combination, in much the same way like electrical resistor's color ring (dot, band) marking; or they have barcode or other encoding scheme avail the device to read them.

2 The device's user interface (touchscreen, keypad, button, speech recognition) allows the patient and nurse to supply either or all of {D, R, T} information.

In very special situations, the device would also allow the entering of refractive index value other than above. This much be possible when, for example, for protecting particular drugs from sunlight the IV sets contain special additives which alter the refractive index of the material (like for UV-resistant IV sets, but not limited to).

Mechanical infusion pumps already have "IV set library". Because IV sets from different manufacturers are made from different material, have different resiliency, their volumetric administration speeds would be different if pumps treat them indiscriminately. So some pumps attempted to be "versatile" by allowing the user to tell it which IV set it is using, and changes motor speed and pressure accordingly. However, pumps only interact (press) with the about 3mm thin very thin tube and it never encloses the drip chamber inside them, so they never ask for {D, R, T}. Therefore it should be clearly that their IV set "resiliency" and "nominal volumetric administration speed" input ability is not a prior art for our {D, R, T} (not necessarily all of them; D is usually unnecessary except the manufacturer might acknowledge that their product's drip mouth might deviate from drip chamber's axis for over a tolerance threshold) user input functionality.

Optical correction

With {D, R, T}, our product in the field could use the information in Fig. A.2.4-5 to Fig. A.2.4-7 to correct the droplet measurement result in two ways: 1. Scaling (multiplication): The eight surface plots in Fig.A.2.4-6 can be combined into a 3D plot, and for any given {D, R, T} (D for the distance between the droplet's mid-plane to the inner surface of the drip chamber, R for drip chamber's inner radius, T for drip chamber's thickness), the ratio of V(D, R, T) (volume in pixels³) value for a standard simulated droplet (either spherical or more realistic pendant shape) to V(D₀, R₀, T₀), an {D₀, R₀, T₀} is a coordinate position where we know how V in pixel³ corresponds to volume in physical units. So the system

V(D, R, T)

always compute V(D, R, T) first, and use the scaling factor contained in Fig. A.2.4-5 to

HD₀, R₀, T₀)

Fig. A.2.4-7 (obtained from simulation of his own system) to get V(D_Q, ^, T^) , and then

V(D₀, R₀, T₀)→ ^(physical) .

2. Another more versatile approach (Fig. A.2.6-4) is first convert from the sensor to image to original image,

Image sensor)→ Image(ob)ect) , by the reverse mapping

Image(ob) ect)— > Image(sensor)] . The mapping between images is essentially the mapping between points, so the system might either

a. Store a set of pre-computed Image (sensor)→ Image(ob) ect) information for different {D, R,

T}, and optionally does interpolation for {D, R, T} combinations that were not computed; b. Do all the computation online. The essential part of the ray -tracing program for our cylindrical surface (customized Zemax DLL files, see section A. 2.2) is less than 100 lines of C code, and will be less than 1000 lines for our lens system with 8 surfaces in Fig.A.2.3-1, together with the code necessary for computing the reverse transfer the size would also be very small and can run efficiently on an embedded processor (for just once during the course of monitoring a complete IV dripping) We can either calculate a conversion matrix, multiplies with the sensor image to get original object image; or more sophisticatedly compute the inverse point spread function, and convolves with sensor image to reconstruct the original object image. Of course, we will then measure the droplet from the reconstructed

V(D, R, T) images. Basically, with the approach the processor does not read pre-computed values

V(D₀, R₀, T₀) but traces rays for each specific combination of {D, R, T} to get the point-to-point

Image(ob)ect)→ Image (sensor)] ¹ relationship, more explicitly expressed as i¾int(object) — i¾e/(sensor) , and compute volume from

Shape(ob ect) = ^ Point{ob) ect) .

Clearly, the second embodiment has greater flexibility in being able to reconstruct Shape(ob) ect) more accurately even for droplets of different diameter, and even different positions (not centered at the axis, or even quite far from the axis). However without any intention to be limiting, we recommend practical implementation to always choose images in the sequence (video) with droplet center closet to the optical axis for better accuracy, and this is achievable because the falling droplets is only at low speeds near the drip mouth location (See US12804163 IV Monitoring by Video and Image Processing, paragraph [88, 89, 90, 107]). The online (dynamic) reconstructing method also works for the pendant, elongated droplet shapes shaped by to gravity and surface tension. We didn't use pendant shapes in illustrations above because spheres/circles were already enough for illustrating the problem (the variation of V according to D, R and T) and the all the inventive elements above applies equally well to pendant droplet shapes.

A.3. Bedside / Ha dheld Monitor

Fig.A.3-1(1) shows a design example from out App. US13897578 and US13903924. The center of the UI design shows the image of a drip chamber, and when it is running (Fig.A.3-1(2)) the drawing is replaced by the video of the dripping process. This LCD screen is mounted on the housing shown in the upper right of Fig.A.3-2 .

We maintain that being able to see the actual dripping speed is important to our user experience. Without seeing the actual drip speed and size user (patient, nurse) will be skeptical to what is going on inside the enclosed housing, since for ensuring a noise-free camera shooting environment we have to use non-parent material to almost completely cover the drip chamber from the outside. The screen renders the internal video outside to the user so they will gain assurance that the device is indeed running according to the numeric figures displayed outside.

However, in experimenting with the device we met the problem that patients have difficulty in accessing the device. This is because the our monitoring and controlling device has to enclose the drip chamber so it is as high as the drip chamber, but drip chamber are often made by IV set manufacturer to the close to the IV solution end, because the higher the drip chamber location, the more convenient for nurses to see it from a relatively far distance. Especially in IV (intravenous therapy) clinics in developing countries, there are usually over 100 patients receive IV administration in the IV clinic room, and it is an established standard to make IV sets with high drip chamber so that they could even be seen from the nursing station in the large IV clinic room.

The IV solution bag, shown next to the IV set in Fig. A.3-2, frequently hangs over 2m high to achieve a higher pressure (atmospheric pressure + PiiquidS¹^¹ the easier fluids can enter the veins, particularly for patients with high blood pressures.

But if we put the device at such a height (above 2m) at the same location with the drip chamber, the patient cannot see the screen' s display, nor can they operate it. For clinic room patients they have to stand up just to see the device display, and they may not be to do so due to weakened strength; for patients on bed the distance to them is often over 1.6 meters, and it is impossible for them (often after operations) to get up merely to see and operate this device.

Therefore our another inventive element is a remote control at the patient side. The remote control has two defining characteristics:

1 It displays dripping video and monitoring information (dripping speed, volumetric speed) on the screen.

2 It is wirelessly linked with the monitor/controller at the drip chamber location.

These two are the basic element enabling the patient to see the dripping process without difficulty. Optionally, it could also have buttons or touchscreen to allow changing dripping control parameters, and use a reverse wireless link to transfer the commands back to the device with the drip chamber. The wireless operating frequency is preferably in the standard 2.4GHz ISM (Industrial, Scientific and Medical), and several transmitters from Texas Instruments can operate at as high as 5000Kbps (625K bytes) rate. When transferring video from device at the drip chamber location to the bed/patient-side display, because the content of the video is simple and repeated, very strong video compression rate can be achieved, and we only need to transfer the compressed information via the link. In addition, the only changing content in the video is the vertical band corresponds to the path of the droplet, so we can only transmit the compressed (and optionally sub-sampled) information of that area, whereas the video of the "background" can be transmitted upfront and cached by the bed/patient side display .

In the practice turns out that the patient/nurse frequently needs to operate the device from the bedside monitor, more frequently than the drip chamber enclosing device, then we can also implement the majority of the electronics and software inside the bedside monitor, and if the wireless transmission rate is sufficient, we may even transmit high- resolution video (only of a small area on the drip's path) to the bedside monitor and do the majority of computation on that. In this implementation the bedside monitor essentially becomes the master device.

To our knowledge there is no prior art for this inventive element. Infusion pumps are placed at low locations, often just the bedside, and they do not enclose drip chamber so no pumps have a video display for dripping video either at the "master" location as shown in the upper-right of Fig.A.3-2, or a bedside or handheld device as disclosed above; infrared ray drip counters never fully enclose the drip chamber so the drip chamber can still be seen clearly, and no has never been video display, either with-device "master" or a remote monitor; the simplest type of monitor, i.e. the solution bag weighter, is essentially a spring balance hanging over the IV solution bag, far apart from the drip chamber. And when infrared drop counter or simple weighter are connected with a monitoring device at the nurse station, they transfer only drop counts/rates, remaining liquid weight, but never video of the dripping process.

The inventive elements in our application can also be used to improve simple infrared drop counters. These devices (Fig.A.4-1) have long been used as cheap monitoring aids, but are inaccurate due to the following problems:

1 They never enclose the drip chamber, and most frequently uses a partial ring to wrap around the drip chamber. As a result they suffer from son (and even incandescent lights) radiation interference.

2 Their drop counting accuracy is significantly influenced by splash/condensation droplets. When droplets on the surface completely block the path of the ray, these counters could no longer detect falling drips (because in their "view" the rays have already been blocked), and fails completely.

3 When the drip chamber is in an oblique (not upright) direction, falling drips could deviate from the detecting rays in which case they also miss counts.

All these problems can be by combing infrared drop counter with inventive elements in this application, as shown in the lower part of Fig.A.4-1 :

1 We can enclose IR drop counter inside the housing as shown in Fig.A.4-1. This blocks external light radiation and solves problem 1 above. The dripping video display element, either the "master" or the bedside "slave", would show the dripping process to the user. 2 We can combine IR drop counter with heating elements in (direct contact, IR heating, incandescent heating) this application to solve problem 2.

3 We can add an accelerometer to so that the processor knows whether deviation from upright direction could cause droplets to be missed, and could alarm the user or take other measures. However when the IR drop counter is combined with a video monitoring camera, the orientation can also be detected by analyzing the liquid surface (level or not). In both ways we would be aware of the event.

We are describing these three improvements (they can be applied independently) to the IR drop counter because it is still has important utility in some situations. In practice not all IV administration requires volumetric measurement and control, for example with ordinary drugs for male adults with small ailments, and these situations are actually very frequent. IR drop counter is attractive because it requires very little power by itself, and only needs a microcontroller (such as the TI MSP430 series) as the processor, and these microprocessors are as small as only 4mm in each side, and consumes less than 1mA or power. So if we add IR drop counters to our precise video-based measurement devices, they could add versatility for non-critical cases and extends the battery life by several times, since the processor is no longer involved with the intense image processing algorithms, but merely routes the video to the LCD controller.

Part ϋ Continuation eo. Condensation & Splash prevention

The (original) purpose of this invention is to create a device that could achieve accurate volumetric measurement of falling droplets during an IV administration, and it works by video and image processing techniques. In this application in we show that a single type of light source does not meet all the challenges in such a system, and propose the combination of an LED (and other higher luminous efficacy sources) and an incandescent light to meet both the illumination and condensation prevention requirements.

This is a relatively new field. Except from applications from the present applicant himself, prior arts include:

US5700692 Flow Sorter With Video-Regulated Droplet Spacing, by Richard G. Sweet. Particularly its Fig.3 teaches the method to compute droplet volume by adding horizontal slices. We acknowledge the existence of this patent (filed 1994) and are making use of their basic methods (adding horizontal slices).

CN 200710168672.3, Enmin Song, A medical infusion speed monitoring and controlling system, which contains very generic description of a system also based on image processing.

US12791885, Ting- Yuan Cheng, "Intravenous Drip Monitoring Method And Related Intravenous Drip Monitoring System".

PCT/US2012/071142, by Deka Research. This applications mentioned incandescent light in an enumeration of light sources, but didn't recognize its function in preventing condensation. Brief Description of Some Drawings

Fig.B. l shows condensation/splash droplet effect.

Fig.B .2-1 shows two resistive heating elements placed behind drip chamber, whereas the camera faces the drip chamber in the front side. The lower heater has a reflector behind it to redirect more light to the drip chamber, and this function can also be assumed by a lens in its front, although not explicitly drawn. The upper heater in the figure has no such reflector, and our intent here is that the reflector/lens is optional. Fig.2-2 shows an example of thermal simulation result using Solidworks' Flow Simulation and we have discussed much of its background in PCT/IB2013/056090 (its Fig.1.3.4-6(2), incorporated as Fig.A.1.3.4-6(2) in this application).

Fig.B .3 's upper part is a model of the air chamber, and the upper graph is part of the drip chamber's internal humid gas mixture's HITRAN (High Resolution Transmission, Sii1 ://www.ci^';).hafvard.edii hi;raii ) absorption spectrum, please refer to PCT/IB2013/056090 for its detail.

Fig.B.4-1 is 0.5mm thick PVC drip chamber's absorption spectrum between 200nm and Ι . ΐμπι, Fig.B.4-2 is the same between l .C^m and 2.5μπι, Fig.B.4-3 is same between 2.5 and 25μπι, and Fig.B.4-4 is a concatenation of the previous three between 0 (or 200nm, lower than 200nm the values are regarded as zero) and 3.7μπι. The 3.7μπι is the typical maximum cutoff wavelength of quartz glass's transmission window, since no radiation from resistive element above 3.7um goes out of quartz glass, in Fig.B.4-4 we do not show values at wavelengths beyond 3.7μπι. Fig.B.4-5 shows a 4.56mmx4mm "patch" window on a PVC drip chamber used to simulate radiation absorption by the PVC. Please refer to §1.2.8 *Numerical Simulation of the 056090 application for detailed discussion.

Fig.B.5's sub-figures are properties of tungsten. Fig.B.5-1 is tungsten's spectral emissivity table from [V. D. Dmitriev and G. K. Kholopov, Radiant Emissivity Of Tungsten In The Infrared Region of The Spectrum, published in Zhurnal Prikladnoi Spektroskopii, V o l.2, No. 6, pp. 481-488, 1965], we have compared them with other authoritative sources and have found consistent agreements. The actual emission spectrum for nine temperatures (Dmitriev's table) are plotted in Fig.B.5 -3, and we will used these data, together with glass transmission window and PVC absorption spectrum, to calculate accurate PVC drip chamber energy absorption. In some calculations in later sections we also use spectral emissivity of tungsten at 2600K which is not shown in Dmitriev's table, and this data (2600K) is taken from Table 2-1 of [Anatoly Pravilov - Radiometry in Modern Scientific Experiments] .

Fig.B.6-1 is Photopic and Scotopic Spectral Luminous Efficiency Functions, taken from volume II, chapter 34 of OSA's Handbook of Optics, and Fig.B.6-2 are the plots. Fig.B.6-3 is the spectrum response of Omnivision's OV7610/7710 sensor. Several later discussions are based on the 5 -million-pixel OV5640 sensor used in our implementation but its spectrum response is not available, so OV7610/7710 graph is used for illustration. We have compared over 12 CMOS sensors from different manufactures and find their spectrum responses differs very little, in fact they all mimic the photopic curve of the human eye. Description of Part B

In this application we describe illumination for an optical IV monitoring device, and a unique problem making it distinct from other machine vision applications is the existence of condensation (and splash) droplets, and the illumination system must have the ability to prevent the formation of condensation droplets in order to acquire clear images. Experiments and calculations will show that it is very difficult to complete the task with illumination source(s) of a single type, and light sources of two types, preferably LED and incandescent, needs to be used together to complement each other.

In a previous application PCT7IB2013/056090, methods for heating the drip chamber have been described in great detail with numerical simulations using computational flow dynamics (CFD) software. Some of the discussions below relates to contents in that application.

B.l. Requirement on Luminous Flax

Please first have a look at Fig.B.7-2: this is an image taken at 60fps frame rate for a falling droplet. The droplet has apparently been stretched obliquely, and this is the result of the default "rolling shutter" mode of the OV5640 sensor we used. The rolling-shutter effect is more commonly associated with images of propeller, sport vehicles and other high speed vehicles, and it appears here for the relatively much slower moving droplet because we were doing close- distance (in an infusion device) macro photography, so even a small distance the droplet moves is equivalent to a large number of pixels in the sensor array.

The stretching effect is overcome in Fig.B.7-2 with sensor's "frame exposure" mode turning, which exposes the entire array of pixels to illumination at the same instant rather than one row after another. The droplet profile is symmetric and its volume can be calculated by adding up volumes of the slices of circular discs.

We may calculate the exposure time based on dimensions and speed of the droplet and measurement accuracy requirements. The volume of each droplet is typically 0.05mL (20 drips/mL), and assuming spherical shape, its radius is

Os< ¾J* 2 . 23538 and diameter 4.57 mm. We would require that during the exposure time the shape cannot be stretched to more than 102% of the actual shape, and when the droplet's image is captured when it is 1cm below the dripping mouth (refer to left sub-figure of Fig.B.l) , assuming free fall, the droplet's speed is V 2 * 3.8 s 0, 01 * 1SGG

To limit the displacement within 2% of the droplet diameter, the exposure time must be smaller than

. _2_.__2_8__5_4__*__0__._0__.2 ~¾

442.7188

- 103.2439^3

With 103μβ exposure time, 250mA is the minimum LED backlighting current to produce image with acceptable brightness and S/N ratio. Please note the darker corners/edges in Fig.B.7-3 : any further reduction of LED power would cause the image to degrade to an extent of impairing volume measurement accuracy.

Can we increase exposure time in order to reduce LED current? Although image would be stretched longer, we can

1 later multiply the measured volume with a scaling factor to compensate for the stretching, such as by or

102%

1

and this would probably recover the actual volume.

104%

This idea is discouraged by looking at Fig.B.7-4: when the droplet' s falling speed becomes faster, the internal interaction of liquid molecules under the acceleration of gravity would alter the its shape, the molecules with lower speed will be pushed or pulled by those falling faster, and the droplet's shape could deviate significantly from a sphere or ellipsoid. When such a shape is further stretched due to extended exposure time, no post-processing algorithm could accurately restore the original shape.

In addition, the droplet's speed of 442mm/s was calculated when assuming its speed is zero when breaking off from the dripping mouth, this only works for very low speed dripping. With moderate or higher dripping speed, droplets would have nonzero initial speed when breaking off from the dripping mouth.

In view of all these, we see that ~ 103 μβ exposure time was in fact close to the limit of exposure time length. In practice we use much shorter exposure time and larger strobing LED power and current, but images discussed in this application were all consistently taken using 103μβ exposure time.

The following table gives technical parameters of the image sensor and LED source:

Sensor

sensitivity 600 mV/Lux-sec

Dynamic range 68 dB at 8x gain

LED

type OSLON LT CP7P,

backlight

a very fine sandblasted glass is placed in front of it to produce

uniform backlighting, and a lens collimator is used to project

light in forward direction

Luminous efficiency 88 1m/W

Current 250mA

Luminous Flux ~ 75 lumen

Calculated using "Relative Luminous Flux" graph in

datasheet, and the absolute luminous flux value at I_F=350mA

Backlight area 314 mm² (diameter 20mm disc)

Lux 75 lm / 314mm² = 238732 lux

Table.B-1

The sensitivity and dynamic range of the image sensor, and the luminous efficacy of the LED, both reflect the state-of- art of their respective technology. If we use shorter exposure time, the current might still increase; but for stretching effect discussed above, the exposure time can hardly increase, and the current cannot be reduced. The approximately 75 lumen luminous flux is the minimum value we need for the backlight.

Luminous Efficacy

Although in the example above we used LED as the light source, there was no intent that LED is the only choice. Other types of light sources with high luminous efficacy, including but not limited to arc lamp, fluorescent lamp, gas discharge lamp, high-intensity discharge lamp (HID), as long as they can be successfully miniaturized and meet safety requirements, can all be used instead of LED.

On the other hand, we don't expect the function of LED to be replaced by an incandescent light, which is due to its extremely low luminous efficacy. An incandescent light at 2600K has luminous efficacy 10.02 lm/W (please refer to Fig.B .9-2 and its respective section for calculation detail; incandescent light efficacy can also be found from many public sources), and

^{75 lm} 7.48 W

10.02 lm/W power would be needed. The bulb size is limited by dimensions of the infusion device as well as optical requirements, and any small incandescent bulb of over 7.48W power rating easily heats up to over 100°C or even 200°C, which if not

7.48 W „_Λ„ .

contained behind shielding would a hazard in a medical device. Also, = 2.02 A , this

3.7 V(Lithium batter)

high current draw would quickly depletes battery and make the device hardly usable.

Another reason preventing us from using incandescent light alone, or as primary source, to heat the drip chamber is that this approximately 7.48 W energy contains over 90% of its energy in the infrared, and according to our calculations in Fig.B.9-1, between lwatt and 2watts of energy is going to be absorbed by the drip chamber. This energy is so strong that the PVC drip chamber would even got melted.

The conclusion we draw from the above analysis is that clearly the PRIMARY light source needs to be one with high luminous efficacy, preferably LED, but other types mentioned above also possible, however CANNOT be incandescent light.

B.2. Condensation

There are problems LED not good at solve. Please look at Fig.B .1 : the inner surface of the drip chamber has dozens of condensation droplets on it, under illumination their shape act as micro-lens and redirects light, resulting in very significant distortion of the droplet image. Because the drip chamber forms an enclosed structure, the evaporated liquid quickly reaches 100% humidity saturation and starts to condense, and it happens all the time, under all room temperature, and for all types of drip chambers.

From physics we know that in order to prevent condensation on a surface, one way is to keep it temperature higher than the gas vapor. This leads us to the idea of heating the drip chamber by illumination source's radiation. The validity of this approach is based on the establishment crucial inequality:

AT (chamber) > AT(gas) this is not trivial particularly because gas mixture of H₂0, C0₂, N₂ and 0₂ has very complicated radiation absorption spectrum, and in fact in PCT/IB2013/056090 we used 4172592 distinct spectral lines (§1.2.1 Gas mixture absorption spectrum) from HITRAN (High Resolution Transmission) data from Harvard-Smithsonian Center for Astrophysics to do the energy absorption integration, and used Computational Fluid Dynamics (CFD) software Solidworks Flow Simulation 2012 to simulate temperature distribution when both the drip chamber and the gas mixture are exposed to the radiation. An illustrational model of the HITRAN gas chamber is shown in Fig.B.3-1 and its absorption spectrum between 0.7 and 0.82μπι is shown in Fig.B.3-2. The calculation procedure and results have been described great detail in the 056090 application for an LED emitter whose peak wavelength is at 3 8μπι, and for incandescent lamps. For LED's whose peak wavelength is in the visible spectrum, we have also conducted the calculation and simulation and have verified the validity of the inequality above. Therefore, LED naturally stands out as a candidate of the solution: heating the drip chamber using LED.

The problem with LED is that its efficiency is extremely low. The transmission window between [0.2μπι, 0.8μπι] of 0.5mm-thick PVC drip chamber, measured at Nanjing University of Science and Technology (¾ SI^^Mlii Φ >ti), using Thermo Nicolet EV220 FTIR spectrometer, is shown in Fig.B.4-1 for two samples (upper; lower is their average), and in the visible band only 15% is absorbed (after correcting for tolerance due to sample mounting, dirt in sample and instrument, and the portion due to reflection, the actual absorption is about 3.5% ~ 5.5%, we can take 4% as nominal). Recall that in Table.B-1 the luminous efficacy of the LED is 88 lm/W, converted to energy efficiency is

88 lm/W

12.88% , use nominal 4% above, the result is only 0.5152%; please also note that this for single

683 lm/W

slice sample, with drip chamber both sides can absorb passed radiation and doesn't differ significant in absorption in that the 2^nd surface absorbs some 80~90% of the 1^st surface, since most energy still penetrate the 1^st surface). The temperature of the saturated gas mixture is dominantly determined by temperature of the liquid, which can be affected by temperature of liquid in the IV solution bag (which might subject to external influence, might be heated, might be under direct sun-light, etc.), by mechanical heating due to the peristaltic pump (infusion pump; our optical monitoring system can also be coupled with an IV pump), so the heating from the LED needs to heat the drip chamber reliably to at least several degrees above the ambient temperature. With 250mA current the maximum theoretical energy absorption

75 lm

from the LED by the drip chamber is only X 4% = 0.00439 W = A.3>9 mW S assuming all ray

683 lm / W

incident on drip chamber at 0° angle, without considering that a larger portion of energy is scatter in other directions, and even the rays incident on the drip chamber will have a larger portion reflected due to Fresnel losses. Therefore, for an 250mA LED the energy a drip chamber (the entire half side) can absorb from it is actually only a few m W, and from our calculation results in §1.2.8 ^Numerical Simulation and §1.3.4 Results of the PCT/IB2013/056090 application, this energy density over the half size of drip chamber area could not cause any material temperature rise. Heating the drip chamber to even the lowest condensation-free (guaranteed, because we need to keep a margin since solution might hotter than chamber, for example if it is taken from another room) temperature would require an inordinately large LED current (several times of 250mA, under »3.3V forward voltage, 2W or more). Just as trying to illuminate the drip chamber with incandescent light is extremely inefficient (7.48W), so it is if we attempt to heat the drip chamber with an LED.

On the other hand, by looking at the absorption spectrum of 0.5mm-thick PVC drip chamber material in Fig.B.4-1 to Fig.B .4-4, we find that its absorption in the IR range is much stronger than in visible range, and the particular characteristics of incandescent is that it typically emits over 90% of radiation in the IR band, so this coincidence prompts us to consider the possibility of heating the drip chamber by an incandescent light source.

Note on absorbance calculation The 4%, or 3.5-5.5% absorbance was calculated assuming some 87%~89% transmission rate for 500~600nm wavelength. The total reflectance is due to both the front film surface reflection as well as multiple (an infinite series) reflections from the 2^nd surface. The total reflectance calculation was consistent with [Hitachi - Measurement of Optical Characteristic of Plastic by UH4150 Spectrophotometer, bttp:// w'.bitacbi- iiitec.c(:'m'^'gioba /scieace/iiv_iii vis/pdf ii3:t4150 datal ^.pdH's PVC measurement (the influence of sample thickness is small on total reflectance). The transmittance data listed in Fig.B.14 is for one of the datasets and in Fig.B.4-1 we see different dataset differ slightly in transmittance. However, some 2-3% difference in transmittance measurement (due to sample/device imperfection, preparation work, etc., and in transmittance measurement this small fraction is with tolerance) might result in change of absorbance with the same magnitude, which relatively is large (we might calculate absorbance as much as 7%). As we discuss in ^Increasing absorbance, the most authoritative should always be obtained from administrations and accredited laboratories and if such data turns out to be more accurate, they should replace our data (although we have already made our greatest effort in ensuring their validity), and the associated definition of heat absorbing" drip chamber should also be adjusted accordingly.

The calculation is based on tungsten's temperature-dependent spectral emissivity data compiled from

1. [Table-4, V. D. Dmitriev and G. K. Kholopov, Radiant Emissivity Of Tungsten In The Infrared Region of The Spectrum, published in Zhurnal Prikladnoi Spektroskopii, V o 1.2, No. 6, pp.481-488, 1965, for 1224 - 2441 K]

2. [Table 2-1, Anatoly Pravilov - Radiometry in Modern Scientific Experiments, for 2600 K]

The emission spectrum until 9μπι is placed in Fig.B.5-2, and is first multiply a rectangular glass (this is very realistic because bulb glass for small lamps is thin, and in transmission bands the transmission rate is over 95%, typically 97- 98%) transmission window, then integrate with the PVC absorption spectrum in Fig.B.4. The percentage of energy absorption is shown in Fig.B.9-1, with row headings denote glass window cutoff wavelength in micron, and column headings denote filament temperature. We see that for window cutoff wavelength over 3.1 micron, the absorption rate

PVC absorption

(defined as ) is consistently over 26%, comparing with the 1.93%

filament emission

PVC absorption

( ) figure, this is such a significant efficiency increase total LED power(= current x voltage)

recommending itself as the ideal heating source!

Note: Fig.B.9-4 plots the data in Fig.B.9-1 and Fig.B.9-2.

BA Combination

Based on the calculations above, we identify the relative merits of LED and incandescent light as: LED Incandescent light

Illumination Superb Poor

Heating PVC Poor Superb so the idea is to team them up, not single type. One (or a group of) LED, and one (or a group of) incandescent. Of course they should optionally be supported by reflectors, lens, or other energy directing device to increase efficiency.

We give a more formal description of our idea:

Behind the viewing plane of the camera, using one (or more) LED, together with one (or more) incandescent light jointly as the light sources, with the LED as the primary imaging illuminator, and incandescent as the primary PVC drip chamber heater.

And we would like to make it again that the word "LED" here is meant to be a generic representative of light sources that are more luminous efficient (in regard to luminous efficacy) than incandescent light, including light-emitting diodes, arc lamps, fluorescent lamps, gas discharge lamps, high-intensity discharge lamp (HID) and so on, excluding incandescent light itself.

With such a combination, the heating power of the incandescent light travels from the back side of the drip chamber to the front side, heating the both sides with power differs very little. This is because of the Lambert-Beer law of medium absorption dl_

m -al

-al where / represents beam intensity, and / is the travelling length in medium. Therefore, if the first surface absorbs 26% of the radiation, the second surface will absorb (1-26%)^χ26%=19.24% absorption, ignoring intermediate gas absorption. This tells us that the back surface will not block the front (viewed by the camera) surface from being heated, and the incandescent light(s) behind the viewing plane can prevent condensation on both sides.

Non-obviousness

No patent/application from other inventors mentioned the use of incandescent light in an IV monitoring system except PCT/US2012/071142 Apparatus For Controlling Fluid Flow . It alluded to incandescent light amongst an enumeration list:

"The uniform backlight 79 may be an array of light-emitting diodes ("LEDs") having the same or different colors, a light bulb, a window to receive ambient light, an incandescent light, and the like. In some embodiments, the uniform backlight 79 may include one or more point-source lights." Yet it has 16 entries of "condensation", at least 14 of which directed toward the image processing removal/noise reduction on condensation droplets. It is clear that the inventor didn't realize the utility of incandescent light as a heating tool, and the reverse side if its visible spectrum radiation is raised to the minimum requirement (-75 lm, see Table.B-1) its IR radiation would cause the drip chamber to melt. A close examination of the 071142 application would further find that it didn't include any content on absorption spectrum, emissivity , or thermodynamics.

Our invention, namely, the particular combination of one (or more) LED and one (or more) incandescent, is only discovered after we have

• built the real device and ascertained the minimum luminous power of the backlight

• measured absorption spectrum of PVC drip chamber material

• using Computational Fluid Dynamics (CFD) software Solidworks Flow Simulation 2012 to simulate

temperature distribution when both the drip chamber and the gas mixture are exposed to the radiation, and proved the important inequality AT (chamber) > AT(gas)

• Obtained temperature-dependent tungsten spectral emissivity data, and calculated its percentage of radiation absorbed by PVC drip chamber

It is the result of a chain of laborious and prudent work, and the invention's beneficial result should not be regarded as obvious by hindsight.

In this section we attempt to further restrict the scope of the invention by giving numerical range on the combination.

First we define absolute radiation power from the incandescent light(s) as absorbed by the PVC drip chamber. In Fig.B.10-1 we defined a rectangular "patch" area of size 10> 15mm² at the backside, and used the same energy generate rate per area as in PCT/IB2013/056090 §1.2.8 ^Numerical Simulation or added new values that the 056090 application didn't include, and these energy generated from the "patch" are used to simulate energy absorption from the incandescent light source. Then we simulated the drip chamber and air temperature using the same techniques disclosed in the 056090 application, and summarize the results, particularly the maximum temperature of found over the patch, in the following table.

Please note that the simulation used fixed heat transfer coefficient 2.9W/K m² as discussed in § Calculation without fixed Heat Transfer Coefficient. In Table.B-2(2) it was replaced by letting the software dynamically calculate heat transfer with Naiver-Stokers equations. Results from both sets were reported.

2mW 16.4474 0.1096 301.57K

4mW 32.8947 0.2193 305.12K

6mW 49.3421 0.3289 308.61K

8mW 65.7895 0.4386 312.06 K

16mW 131.579 0.8772 325.36 K

30mW 246.711 1.6447 347.36 K

40mW 328.947 2.193 362.47 K

48mW 394.737 2.6316 374.3 K

60mW [2] 493.421 3.2895 391.71 K

72mW [2] 592.105 3.9474 408.80 K

80mW [2] 657.895 4.386 420.05 K

Tabled

^[11PVC's melting point spans a wide. Please refer to [Charles E. Wilkes, PVC Handbook] . States associated with large temperature values should be treated with caution. not included in Fig.B.13(l)

Please note that simulation could only approximate the reality. Although we made our best effort to enter accurate PVC drip chamber material properties, these properties of polymer plastic vary with temperature and engineering handbooks does not list all values.

We identify four particular patch point highest temperatures, 312.06, 325.36, 347.36 and 374.3K and their associated energy generating rate, they can all be served as "endpoints" in the simulation results (and possibly in patent claim discussions). We believe 374.3K (101.15°C) cannot be exceeded and this is one of the possible upper bounds (and its associated parameters) for the radiation absorption from incandescent light's radiation over a 10* 15m² area.

374.3K→2.6316 mW/mm²→2631.6W/m²→(times 683 lm/W)→1797382.8 lux

§Area Definition

First we pick 2.6316 mW/mm² as the maximum luminous flux per area over the 10^x15mm² area on PVC drip chamber from an incandescent light, with PVC material 0.5mm thick (the most typical thickness of PVC drip chambers). This is an absolute limit, and for different PVC thickness and radiation area smaller or larger than the 10* 15mm² area, it should all be adjusted accordinglv(this requires a definition of the area. For example, we may define the "area" to include area where the radiometric/photometric incident power level (e.g. watt/mm^A2. lux, etc.) is no less than 5% (or/can elect 10%, 20%, 30%, 50%)) of the peak level.

In Table.B-1, the LED's lux is 238732 lux. The 1797382.8 lux looks larger than LED, but it is an upper bound for PVC's absorption (to prevent health hazard, melting, etc.), and because incandescent light's energy proportion in visible spectrum is low, we need to calculate a range for its photopic lux over the 10* 15mm² area. For different filament temperatures and (the filament cannot be too high such as over 3000K for service life consideration, but 2600K is the largest temperature for which we could find relatively complete spectral emissivity data) glass transmission window cutoff wavelengths, the percentage of photopic radiation among total radiation power is calculated in Fig.B .9-2, which does not change across cutoff wavelength since no window we simulated shrinks into visible spectrum.

The ratio between PVC's absorption and photopic radiation is calculated in Fig.B.9-3, where we see that the smallest ratio 6.69082 is when cutoff wavelength is 1.3μπι and filament temperature is 2600K. Values close to this are clearly obtained by blocking significant portions of the IR radiation (reflected, redirected to filament, absorbed or reflected by coating on/in bulb etc.) .To achieve this, the above mentioned measures (in parenthesis) might not be enough, and we might need to optionally put an IR filter after the incandescent light (before drip chamber, different from the more conventional IR filter in front of the camera), which is also a means to control radiation from the incandescent light(s) to prevent it from heating up the drip chamber to unacceptable temperatures. lux(incandescent light)

Using the 6.69082 ratio, we calculate a theoretical maximum ratio in which we would lux{LED)

like to limit the power of the incandescent light: lux(incandescent light) I 797 82, 8 / 6.69082 25V lux(LED) ^~ 238732

We define this as the relative maximum lux ratio in our LED (in the generic) and incandescent combination.

A difficulty we acknowledge is that the lux of the incandescent light is defined over a 10* 15mm² patch on the drip chamber, whereas the lux of the LED (Table.B-1) were defined over the backlight area. Although lux is an area- averaged quantity, the two areas over which the two measurements were defined do not have an exact 1 : 1 relationship, however there is no better way to define the two quantities. It is also highly possible that the area which the incandescent light illuminates (or heats) is smaller than 10> 15mm² so that our definitions do not directly apply. The positioning of the incandescent light(s) could also be different from the LED backlight, further complicating the case.

Nevertheless, the difficulty in phrasing the exact formal definition should not overweigh the significance of the essential inventive element. The chief contribution of this application is the recognition of the

LED Incandescent light

Illumination Superb Poor

Heating PVC Poor Superb relationship, and the idea of combining the two. It is in this application we first propose this unique combination after built the real device and ascertained the minimum luminous power of the backlight

measured absorption spectrum of PVC drip chamber material • using Computational Fluid Dynamics (CFD) software Solidworks Flow Simulation 2012 to simulate temperature distribution when both the drip chamber and the gas mixture are exposed to the radiation, and proved the important inequality AT (chamber) > AT(gas)

• Obtained temperature-dependent tungsten spectral emissivity data, and calculated its percentage of radiation absorbed by PVC drip chamber

In the fundamental there is large latitude in choosing the ratio of the respective lux, even allowing incandescent light to have higher lux than LED (generic sense) . Rather than trying to be liberal, the calculations above reflects our effort to derive, to restrict, and to recommend relatively more "economical" combinations whose scope constitute only a small portion of the full theoretical range.

Also, in aware of the difficulty with the lux ratio formulation, we propose another quantity, the

Viewing Plane Luminous Flux

, defined by the light source(s) 's luminous flux calculated over the viewing plane, which coincides with the drip chamber's axis (the axis passes the plane) and perpendicular to the optical axis. It calculates the absolute value of flux without considering whether the light source(s) is placed before or after the camera's viewing plane.

With this definition, the 28.13% figure remains unchanged, but its significance changes to

Viewing Plane Luminous ¥l x(incandescent light) $25°/ Viewing Plane Luminous Flux(LED) as the upper limit.

Using the other three values; 31 IK, 347K and 392K, we calculate another three possible upper bounds: 312.06K→0.4386 mW/mm²→438.6W/m²→(times 683 lm/W)→299563.8 lux 325.36K→0.8772 mW/mm²→877.2W/ m²→(times 683 lm/W)→599127.6 lux 347.36K→1.6447 mW/mm²→1644.7W/ m²→(times 683 lm/W)→l 123330.1 lux and three upper bounds (derived from Table.B-2 values), defined either by lux ratio (LED(s) from backlight area, incandescent light(s) on 10^x15mm² patch, or the new Viewing Plane Luminous Flux definition), and the values are

109.6 x 683 / 6.69082

4.6864%

238732

219.3 x 683 / 6.69082

9.37712%

238732

328.9 x 683 / 6.69082

14.0635%

238732 299563.8 / 6.69082

238732

599127.6 / 6.69082

37.5085%

238732

1123330.1 / 6.69082

= 70.3262%

238732

These values are the calculated bound "critical points" used to discern if a resistive radiation source (particularly tungsten filament) is actually a heater or illuminator, and they are calculated assuming the theoretically maximum (PVC absorption : photopic) ratio (if camera not using visible range, use the corresponding "camera- wavelength- photopic" definition instead which is straightforward), and the actual photopic power of the incandescent source is usually much, much smaller the value used in the above calculation since the 6.69082 value was obtained using 1.3μπι cutoff filter for the incandescent source, very much an artificial contrivance. The number, for example 18.7542%, suggest that photopic power from incandescent source is less than one fifth of LED power, which clearly indicate that the inclusion of the incandescent LED is essentially a heater, rather than illuminator. For other thresholds such as 4.6864% such that photonic power from incandescent is less than 5% of LED, its heater nature is beyond discussion. We therefore use value 18.7542%, or allow electing from the above six figures between 4.6864% and 70.3262%, as thresholds for classifying incandescent sources as either "illuminator" or "heater", and "heater" type is what we exclusively claim.

The fundamental reason why we need to use LED and incandescent lamp combination is the simultaneous requirement of illumination and condensation prevention. And in Fig.B .4 we see that PVC actually absorbs more or less in all bands, and LED only fails to heat PVC because it efficacy is low. So what about a higher efficacy source?

This idea opens the gate of a flood of single type and compound (combination) type of light sources. We classify them logically:

First we stipulate that illumination source type counting does not include incandescent light, since its combinations with all other types of lights have been exhausted in this application and in PCT/IB2013/056090.

For other type of light sources: an exemplary of lighting (in this example strobing LED light without intention to be restrictive) control circuit is shown in Fig.B.ll(l) (with Multisim's SPICE simulation, all SPICE models are from manufacturer's website), and the existence of 0.0001Ω R15 is purely for converting current to voltage for displaying purpose. The strobing (and exposure) length needs to be short as having been discussed previously for because droplet traverses large viewing angle in short duration due to the close-range optical nature of the configuration.

With P-Channel MOSFET (or N-Channel, or BJT, or IGBT, or JFET, or numerous type of controlling elements, or numerous of integrated switches, or dedicated LED drivers with current controlling ability (usually via feedback loop adjustment)), voltage (middle) and current (top) over LED follows closely that of the control signal (bottom) with shut- off time less than ΙΟμβ, and this is also what we have measured using real oscilloscope.

In certain cases, the shut-off time of the circuit is influenced by capacitance (including board's parasitic capacitance) and resistance elements and could lead to long shut-off time. However, this could only be a fraction of the frame interval (e.g. ΙΟΟμβ shut-off time VS 25ms frame interval).

First we aggregate the sum of radiation (including illumination) sources on during exposure time, as well as the types of sources on during exposure time, as A; out of exposure time the corresponding aggregation (type and sum) is B.

The luminous level for both type A and type B light sources (which might be same physical source, but differentiated as two, or more types/levels in cases when there are more than two different levels) is defined over their respective operating time:

Because the fundamental flux unit for radiometry is watt, which is energy divided by time, therefore all derived its units is radiometry or photometry are averaged over their respective time interval.

Therefor for example if FPS is 40, exposure time is 200μβ, then non-exposure time is 25ms-20C^s=24.8ms. The luminous/radiant level (over area, or over solid angle, over area-solid angle, or whatever) is defined over their respective time length.

The area over which the sum/aggregation works is either viewing plane, or [an area inside, including or overlapping the viewing window], or a plane between the back radiation source and camera and being parallel to the viewing plane.

In view of the ratio between exposure and out-of-exposure time, it is clear that the lighting provided out of the exposure interval is for heating purposes. The reason we discussed varying-level lighting scheme in priority applications in the context of CMOS sensor is not because it works only for CMOS but not CCD, but because with CCD sensor the drip chamber can also be heated by constantly -on light sources out of the exposure time, out of exposure time the source clearly served as a heater instead of illuminator. Using either varying or constant level lighting scheme intentionally with non-zero output level out of exposure time is clearly because after recognizing the absorption spectrum and heating effect for condensation removal which have been described in our applications. Therefore, we explicitly reiterate again the varying level dichotomy in priority applications to explicitly include cameras with true global shutters, and in addition we would like to expand the (electable) value set of r to {l.O(constant), 1.1, 1.25, 1.5, 2, 3, 4, 6, 10, 15, 20, 30, 50, 75, 100, 200, 500, 1000, 2000, 3000, 5000, 10000, 20000, 50000}(please note that if the ratio is smaller than 1, it clearly suggest the heater nature upon first glance), in which new figures in addition to 120% are added, for both camera types with (e.g. CCD) and without (e.g. CMOS) true global shutter.

Some more elaborations on cameras without true global shutter is necessary because of the cases depicted in

Fig.B.ll(2): the 3^rd row corresponds to an ideal two-level dichotomy and the 1^st row shows zero output level after exposure. However for cases in row 2 and row 4 one might introduce ambiguity in construing when marks the end of exposure. Row 2 shows that one can "lengthen" the pulse so that extra length is chiefly for heating (which also affects image quality), and after the added length the output is reduced to low level (usually near zero, but can also be non- zero); row 4 shows that the shut-off time might be very long (the drawing is exaggerated) so that between exposure time there is appreciable light level.

When in cases like row 2 and row 4, and when exposure time cannot be defined by camera per se, we introduce an alternative definition:

• The exposure time is defined as the time interval between which the output level (defined either by LED ' s voltage, current, radiometric/photometric units, as received by the camera) is > 50% (additional, we propose electing 60%, 70%, 80%, 40%) of the peak level. The output level (during exposure time) is calculated either by peak, averaging, RMS averaging or other reasonable methods.

Note that this 50% definition can also apply to sensors with true global shutter as an alternative definition to its intrinsic exposure/non-exposure definition.

In addition to restrict out-of -exposure-time light output level by its relative ratio with exposure time, we also attempt another (can exist simultaneously) definition with absolute energy level:

• When not in exposure, the monitoring system uses a light source, whose output (defined by 0.5mm-thick PVC absorption) defined over a 10* 15mm^A2 (adjust accordingly with area, see §Area Definition) area in on either the front or back of the drip chamber is lower than 2.193 (can also elect other values in Table.B-2(l), e.g. 1.6447, 0.8772, 0.4386, 0.3289, 0.2193, 0.1096) mW/mm², or corresponding values in Table.B-2(2) (Dynamic heat transfer coefficient) if dynamic heat transfer coefficient is used.

A note on non-obviousness

One might view the use of varying-level output as from the perspective of coding scheme: the use of flashing lights to freeze fast moving objects was established, as well as the use of longer time exposure for relative static objects. However, no prior art has taught the use of radiation source to prevent condensation in a drip chamber, and the validity of this approach is established as after our arduous calculations in this and prior applications. In fact, exhausting prior art would only find teaching on heating the liquid bag or heating the tube near the vein for warming effects (citation needed), but no teaching on heating the drip chamber. In addition, all prior arts whichever mentioned "condensation" suggested its treatment with image-processing techniques, rather than physically preventing it. Therefore, our use of radiation source for condensation prevention is clearly novel and inventive, and the use of varying-level output is in fact one of its implementations, whose discovery was not incidental. The extra length has clearly defined purpose of heating yet at most with only minimum residual illumination effect (in cases of no true global shutter). More computations have been done on a Dell R510 PowerEdge Server with 24 threads and the results are shown in Fig.B.13 (1) to (3). These data are found to be consistent with results in 056090 application (The results in an earlier submission of the same table are found to be larger, probably due to issues in file system unnoticed between transfers).

Further supplemental data are added between Fig.B.15(5) and Fig.B.15 (8) for the full range of possible thermal conductivity values ranging from 0.14 to 0.28 W/(m-K), adding 0.14, 0.24 and 0.28 to the original 0.19W/( m-K) used since 056090 application, and the resolution of computation mesh grids was set to 0.0001m=0.1mm, 1/5 of the minimum drip chamber surface thickness. Comparing with previous data we see the difference is little at low temperature (below 320K, note that 42°C is one of the critical temperature for protein) which suggest that our previous computations has been reliable. For the added 0.14, 0.24 and 0.28 data the mesh resolution are also 0.0001m.

In ^Increasing absorbance, we have made linearity argument and combining with experiment data showing that temperature increases with thickness. This has been verified again for all added data to be correct.

In the table (Fig.B.13(1)), we see that if heat generation rate is low such as only 1 mW, for the small window temperature average is only 1.55K. If the IV solution bag was taken from another environment with higher temperature (such as 5K), then the liquid retain their higher temperature for a significant period of time due to water's high heat capacity. Liquid immediately heats air within drip chamber when infusion starts, so air easily gets to 5K higher than environment whereas drip chamber inner surface is only 1.55K higher, and humid air would quickly form condensation on the inner surface.

We therefore propose radiation power adjustment:

(1) let there be a temperature sensor to measure temperature of the liquid. To distinguish with prior art "liquid warmer" which also has temperature feedback mechanism (like a thermostat), we propose that a. The temperature measurement is done (by whatever temperature sensing device) farther from the

injection/administration site at the human body. More specifically, we limit temperature measurement location to be between (at) IV solution bag and 10cm away from administration (needle) site, and if overlapping with prior art, we further reduce the range, in steps of 2cm, away from the administration site and from the liquid solution bag, in a manner that is not necessarily symmetric (e.g. not that x cm farther away from site implies also an equal x cm farther away from IV solution bag). The temperature sensor can be placed anywhere in this range, but is preferred to be near the drip chamber, such as directly measuring drip chamber's liquid area temperature, or between drip chamber and solution bag. Our prior patent examination experience with SIPO and USPTO shows that there has been no prior device heating the drip chamber, and no such measurement on temperature of the drip chamber (either on its surface not in contact with the liquid, or on its surface in contact with the liquid). b. What makes it absolutely different from prior art is that this temperature is being used to control light output level. Upon finding that the liquid temperature is higher than drip chamber (particularly the viewing window(s), or any of the viewing window(s) in case of multiple camera), the device have ability to adjust i. Radiation power as being absorbed by the drip chamber (raising current, voltage, number of output sources, change wavelength, etc.).

ii. Heat generation rate of contact heaters (disclosed in 056090 application, no prior art have been found during PCT International Search Report; also disclosed earlier in app US 13897578; We have actually alluded to this output power adjustment ability in 056090's §1.2.8 *Numerical

d

Simulation in which we suggested varying— of viewing windows to accommodate different use

dt

cases and environments ).

c. In fact already mentioned in point a. above, we include a sensing/measuring capability to measure

temperature of the viewing window. This can be done be contact sensor (like being a part of the contact heater in 056090) or by thermal imaging, or by analyzing camera image if the camera has such ability. Please note that as we show in ^Refractive Index Correction, the change of plastic (there PVC is measured)'s n can be significant. Therefore we propose two distinct uses of the temperature measurement:

1) Output power/heating temperature adjustment

2) Optical correction whether the acquired viewing window (front and/or back) is used for either purpose can be easily discerned by examining whether the respective actions are performed.

It is further proposed the upon a) measuring and fining, or 2) via thermal calculation (or retrieve parameter-result pairs stored in device's database), that the current output power from contact heater or radiation sources (including illumination light) would heat the drip chamber inner surface to above 42°C (actually we mean any critical temperature for human, for protein, for specific drug; protein is usually stored in lower temperature, other drugs might tolerate higher or lower, and such information can be obtained from user input, label reading (barcode or camera), electronically(USB input via inserted disk (SD card), etc.), wirelessly).

The combined effect is that: we propose a system which dynamically monitors and heats drip chamber (particularly view area) for video monitoring system, and the heating power and/or heating temperature is

limited/adjusted/controlled according to environment, liquid and drug tolerance temperature.

All of our knowledge on the latest patent literature suggests that there has been no prior with radiation/contact heater output power adjustment ability according to liquid and/or drip chamber (particularly viewing window) temperature measurement.

The temperature sensors types and locations, etc., also applies to ^Temperature Sensor used during Contact Cooling/Heating.

Accommodate Drip Chamber material change It is further proposed that the device has the ability to accommodate different drip chamber material. Plasticizers in PVC is considered to have adverse health effects and there are alternative materials such TPE plastic (Thermoplastic elastomer) . In visible range transmission rate are similar for many plastics (near 90%, see Hitachi - Measurement of Optical Characteristic of Plastic by UH4150 Spectrophotometer, http://www.hitachi- hitec.com/global/science/uv_vis/pdf/uh4150_datal_e.pdf) and are only weakly affected by thickness. Transmission rate difference such as between 88% and 90% are difficult to be detected by camera, or if it could, doesn't justify significant adjustment of illumination light level. However, density, heat capacitance, thermal conductivity and absorption rate might differ significantly (see 056090 's §Drip Chamber Materials), and will cause very significant change in heated temperature (ignorable or inordinately high temperature rise). Therefore, upon knowing (by user input, analyzing drip chamber shape, by temperature sensor (described above) feedback such that chamber temperature is not as expected/calculated from thermal model, by observing unexpected condensation, or whatever means) a different material/drip chamber, the output level of part or all radiation sources is adjusted, which is what we sought to protect; the unprotected scope could at most vary radiation source level proportional to material transmission rate change.

Please note that as we have mentioned TPE, and since the Hitachi link included PMMA, PET, PC, sufficiently many example of alternative materials have been given. Since we have disclosed in great physical and mathematical detail method and data for the thermal analysis, all figures, limitations, ratios, concrete values in throughout the entire sequence of application is allowed to extend to other specific types of materials in claims, provided that the same analysis/calculation procedures are followed.

Restriction on absolute drip chamber surface temperature rise (most pertinent should be inner surface, but since inner surface is hard to measure except for in simulation, so we extend this definition to outer surface, and to drip chamber are except for those in directly contact with the lower liquid.)

We propose that the combined output level of radiation sources (plus contact heaters) heat any area of the drip chamber, particularly area that is not in direct contact with water, further particularly the viewing window (take definitions in § Splash-proof drip chamber + Video monitoring Solution), be more than 0.5K higher than either temperature of {environmental air surrounding drip chamber, or liquid in measured by sensors installed in positions mentioned above}. If overlapping with prior art, we would elect any value listed in the range of Fig.B.13 and Fig.B.15, and the 056090 application (as well as in table.B-2's whose data need further verification), including IK, 2K, 3K, 4K, 5K, 6K, 7K, 8K, 10K, 12K, 14K, 16K, 20K, 25K, 30K, 35K, and any of the interpolated values within the range; or if convective heat transfer coefficient is dynamically calculated, use values in Fig.B.15 and Table.B-3 values instead in the same manner.

Why these temperature rises are not incidental effect of the light sources?

All incandescent light sources have been have already been exhausted. In practice LED is preferred because its fast microsecond-speed response time is particularly suitable for flashing, and because its concentrated wavelength eliminates color aberration. Mid-IR and far IR LEDs have very small output power which is not sufficient for illumination. The OSLON LED we use have luminous efficacy of 88 lm/W, or 12.88%, and for 0.5mm thick PVC absorption in these range is about 4%, so only 12.88%^x4%=0.5152% power is absorbed. To heat a 10* 15mm² viewing window's average temperature to be 0.5K up, Fig.B.13(l) right table 9* row shows that 0.5mW is the minimum (for 0.34K rise). 0.5mW/0.5152% = 97mW.

Using dynamic HTC, to rise 0.38K we need lmW, and lmW/0.5152%=194mW.

As discussed in §1. Requirement on Luminous Flux, exposure time is 103 μβ for fps=60, frame interval 16.7ms, and current 250mA. The voltage is 3.22V (see OSLON LT CP7P datasheet), and the actual input power is

250mA^x3.22V^x103/16667 = 4.975mW. Comparing with the requirement to raising 10^x15mm² area average temperatures only 0.34K above which requires 97mW input power, the ratio is 97/4.975=19.497. This means that, we have to raise output level about 20 times greater than that is sufficient to produce clear and bright image to only achieve an 0.34 K (dynamic HTC: 0.38K†, 194mW, 39 times) average temperature rise over the small 10^x15mm² window.

The images in Fig.B.7 (discussed in §1. Requirement on Luminous Flux) were taken with 1.0 gain (no additional gain) which reflect true pixel capacitor storage at this level, hence if we only output above, say, twice or a bit more, we immediately reach overexposure level of the sensor. The electronic imaging process is that first each pixel has a short time of pre-charge, then during exposure incident photons causes discharge, whose result will in next stage be sampled by ADC. This type of overexposure (essentially over-discharge) cannot be corrected by smaller digital gain in later stages.

What temperature rise is enough? There are always uneven temperature distributions even indoor environments and over 2 degree difference in the same room is not uncommon. Temperature variations can also be caused by air conditioner, artificial and sunlight incident on IV solution bags, patient (and entire IV set) being transferred, IV bags taken from a different floor (building) in the hospital, etc. Very commonly a IV bag is taken from preparing room, or another ward room, and liquid retains temperature for a long period of time due to water's high heat capacity (discussed above), and will very quickly cause condensation as air is heated almost immediately to liquid temperature and then condense on the cooler surface.

What is more, once condensation droplets are formed, they cannot be removed by only a few degrees of temperature because of hysteresis effect.

Therefore, at several degrees of "safe margin" must be kept. Refer to Fig.B.13(l), to raise (average) the 10^x15mm² area:

1) 1.38K → 0.002W ÷ 0.5152% = 388.2 mW input, 78.03 times of 4.975mW LED input

2) 2.75K- → 0.004W - ÷^■ 0.5152% = - 776.4 mW input, 156 times

3) 4.1K - >0.006W ÷ 0.5152% = 1064.6 mW input, 234 times

4) 5.34K- →0.008W^■÷ ^■0.5152% = ^! 1553.6 mW input, 312 times

5) 6.08K- →0.012W ÷ 0.5152% = 2329 mW input, 468 times 6) 10.64K→0.016W ÷ 0.5152% = 3106 mW input, 624 times

7) 15.62K→0.024W ÷ 0.5152% = 4658 mW input, 936 times

8) 19.26 K→0.03W ÷ 0.5152% = 5823 mW input, 1170 times

If we increase output light level to these multiples, there exceed overexposure threshold several hundred times, and incur crosstalk or even possible damage to the sensor IC. For the input current alone, the original 250mA current would be raised to some 40 A or even hundreds of ampere, leading to another array of complex circuit problems.

From the above calculation, we see that sources stronger enough for high-quality, instantaneous, distortion-free droplet imaging is on the scale of hundreds weaker than competent heater output level. Therefore, from perspective of the effect on the temperature rise of the viewing window, we claim the scope of any illumination source/heating source/radiation source/contact heater which result in over 1.38K (or elect lower 0.34K, 0.69K, or elect higher 2.75K, 4.1K, 5.34K, 6.08K, 10.64K, 15.62K, 19.26K, 22.82K, 25.16K, 29.8K, any value included in Fig.B.13(l) (other tables for different thickness; Fig.B.15 for dynamic HTC calculation results) which we used as example, or any value from the entire sequence of tables) temperature.

Note that if any compensation is taken such as reducing the gain, it might not work because over-exposure already happens at pixel cell's capacitor storage; if by any post-processing, or ask camera manufacturer to alter photosensitive sensitivity level down from existing product, or filtering (lens coating) wavelength that is used for shooting, to prevent overexposure, that is obviously all secondary "evasions" after making the "primary" use of light for heating.

We therefore define and restrict temperature rise due to illumination sources above levels defined above to distinguish "heater" from "illuminator". And we claim the exclusive use of the "heater" type.

Increasing absorbance

In addition to being plotted in Fig.B.4, transmission percentage data are also listed in Fig.B.14, in lnm increment. There is overlap between [200,1100]nm and [1000,2500]nm because they are measured with different instruments (accessories). When they differ, we stipulate that they should be replaced by more authentic data; when more accurate data not available, we stipulate that by default whichever data has higher transmission, at a specific wavelength, is taken.

Absorbance is calculated following [Hitachi - Measurement of Optical Characteristic of Plastic by UH4150 Spectrophotometer, bltp:/fo- w,bitachi-bitec.coin^^ which is from [JIS

K7375 0 φΙ¾ϋ¾·. JIS K. 7375 : 2008. 7^" 7

Determination of total luminous transmittance and reflectance], or by fundamental optical law calculation. For the OSLON LT CP7P we used, the 0.5152% input power→absorbed power rate is due to the multiplication of low luminous efficacy and PVC's low absorbance at visible range. Some LED achieve over 30% efficacy in near IR range yet the absorbance is not significantly higher. At wavelength longer than 1 μπι, semiconductor emitters with large output power are rare (some has only less than 2mW output for each unit), and if we elect this range it is imperative to use resistive heater (tungsten, etc.) or other types of higher efficacy radiators.

With whatever radiation source, however particularly for LED, it is advantageous to let the drip chamber absorb more. Presumably, this can be done by two ways

1) Increase thickness of chamber material (Lambert law I = e ^l , see 056090 §1.2.2 Absorption spectrum ofPVC). However, mass also increases, and would require more heat for fixed ΔΤ rise than when it was thinner.

To establish this fact (that increasing drip chamber material can increase temperature of the inner surface of the viewing window when it is exposed to the radiation source) we give the following proof:

I. Refer to Fig.B .23, the net amount of reflected, absorbed and transmitted light need to be calculated with infinite series. We use r to denote reflected light at PVC-air interface (note we use PVC here as in calculations elsewhere because it is the type of our measured samples and because it is the most common type, but our methods are applicable to all types of materials; and always assuming 0° incident angle for simplicity), t for transmitted light through surface, p denote one-pass (hence "p") remaining percentage after Lambert-beer absorption through thickness/length L. The total back-reflected energy is back

= r + + (rpf + (rpf +^■■■ (rp)^∞)^■ t

\ - (rpf

The total front-going energy is front

tpt . (\ ₊ (rpf ₊ (rpf ₊ - (rpY)

for PVC

t = \ -r = 0.95742 and the total absorbed energy

A

= 1 - front - back

n_n^An 0.916653» 0.0390311??²

= 0.95742

1.0-0.00181306/ 1.0-0.00181306/

Because of the magnitude of the quadratic term is small, the value is dominated by p and is in fact linear over a wide range. When plotted in Fig.B.24, we see that between [0.75, 1] p is almost perfectly linear and close approximate 0 as p→l.

Therefore A oc \- p

A∞C(\-p) = Ca in which a is the "single-pass" absorption.

II. The relationship of "single-pass" absorption with length L. Because Lambert-Beer law is exponential, hence for

L = cL_n

I- a = (l-«₀) a∞ka.

■^■■a∞L where we used small-value Taylor expansion. Further,

III. Temperature rise vs chamber material thickness (L)

In Fig.B.13(4) (used fixed heat transfer coefficient) and Fig.B.15(4) (dynamically computed heat transfer coefficient) comparisons were made for same energy input (generation rate over the viewing window) - different thickness chamber. We see that although mass increases with increasing L (denoted as T in the figures), temperature drop is fairly small. This suggest us that at steady state the conservation balance is primarily due to the convection loss to counter the energy input, hence the steady temperature of the viewing windows changes only a small magnitude to keep the gradient slope for convective energy dissipation.

The most compelling answer is given by inspecting the tables cells within Fig.B .13 (4) and Fig.B .15 (4) : For example for thickness=0.5mm, if the thickness increases 3 times, because A cc L , energy input also increases ~3 times, therefore we shall look at the table item at column "thickness^ 1.5mm" and row "energy = 3 times previous". In this way in Fig.B.15(4) tuple {0.004W, 0.5mm, 299.43K} maps to {0.012W, 1.5mm, 301. IK}, and {0.02W, 0.5mm, 304.06K} maps to {0.06W, 1.5mm, 311.08K}, in both cases we see a considerable temperature rise, and this is also true for other cells mappings (for at least the majority; the trend is evident).

Data with finer mesh resolution and additional possible thermal conductivity 0.14, 0.14 and 0.28 W/(m K) has been added between Fig.B.15(5) and (8), which also agrees with our conclusion.

To summarize, step I and II proved relation A cc L for relatively thin chamber material, and step III used this relationship together with CFD simulation result to confirm the fact that increased thickness leads to increased temperature rise at the inner surface of the viewing window.

The generality of method extends to all materials, hence we seek to claim first broadly, increasing the thickness of the drip chamber material, either particularly at viewing window (front (facing camera) and/or back) or more uniformly over the drip chamber, to thicker than 0.5mm for increased temperature rise/heating efficiency of the inner surface of the drip chamber.

We may further limit the length to either within the range [0.5, 1] or [0.5, 1.5] or [1.0, 1.5]mm, or pick any range in between [0.5, 1.5], oruse any number within [0.5, 1.5] as the lower limit and specify that drip chamber viewing window area be thicker than it.

We may still further restrict, as in §Splash-proof drip chamber + Video monitoring Solution, for the specific combination/solution of IV video monitoring device, optionally but preferably with condensation prevention capability /mechanism via heating and/or cooling (§Cooling and combine Heating/Cooling), with thickness-increased drip chamber (defined above) for higher energy efficiency. Such a device-accessory combination is intended to avoid potential overlap with existing drip chambers.

Thinner drip chamber/tube for increased cooling efficiency In a similar vein we suggest, in consonance with §Cooling and combine Heating/Cooling, drip chamber with material at cooling sites thinner than 0.5mm, so that cooling mechanism might get more direct contact with the for increased cooling efficiency. The thickness of the thinner drip chamber (or tube area) at the cooling sites (either at the drip chamber area when cooling liquid within the chamber, or at sites above drip chamber when cooling applies to tube length above it) is restricted to be smaller than 0.5mm (or elect from 0.4mm, 0.3mm, 0.2mm, 0.1mm, 500um, 300um, ΙΟΟμηι, and any value in-between).

2) Alter chemical composition of PVC. As [Shimadzu Corp.

of Polyvinyl Chloride, Sii1p://wv.'W.iii;mdei.c;)/ap!) catioii noses/A 42 Lpdfl shows, there are strong absorption peaks due to additive, including plasticizer. This suggest that a host of methods to increase absorbance for a particular radiator. This might include a) Adding structural / physical property additive

b) Special treatment

c) Adding colorants, color absorbents and others.

We therefore propose, as in section §Splash-proof drip chamber + Video monitoring Solution, first as

peripheral/accessory : a. Drip chamber made of material which at least in a 0.3nm range absorbs "significantly more" radiation than drip chamber material with properties shown in Fig.B.4 and Fig.B.14. These drip chambers are from four brands in China: Yong Kang (i¾J$), Kang Ye Da (JftB†¾, Kang You (Mix) and Sheng Guang (¾;¾) and are 0.5mm thick ordinary gravity IV sets, they are registered in China's SFDA and their properties can be obtained from the administration. The narrow 0.3mm continuous range is specified because many light sources have very stable and narrow peak, for example sodium's has a virtually monochromatic 589.3nm peak, actually two dominant spectral lines very closely together at 589.0 and 589.6 nm. We might also elect from {0.1, 0.2, 0.5, 1, 2, 5, 7, 10, 15, 20, 25, 30} for the width of the "window" (as the 0.3 used above). b. The said range has non-zero overlap with a ±35nm range of at least the 5* peak (up include 4^th, ... , 1^st) wavelength of all lights whose radiation ever touches the drip chamber. The peak is ranked by first time averaging (summing) absolute either absolute radiation power, or power as absorbed by drip chamber, over periods longer enough to smoothly cover all different period lengths of sources (in the cases of multiple camera not shooting at the same time). Although 5^th is specified here, in practice this should usually be the 1^st or 2^nd strongest peak (we might elect 1^st or 2^nd or 3^rd or 4* instead of 5^th in claim).

c. "significantly more" means the absorption percentage is at least absolutely 1% higher, or relatively 25% higher (1% is 25% of 4%, and 4% is the value for the ordinary PVC drip chambers cited above in visible and near IR range), than data in Fig.B.4 or Fig.B.14. Alternative values to elect instead of 1% include {2%, 3% 4%, 6%, 8%, 10%, in 1% steps up to 99%, as lower bound of increment}; alternative values to elect instead of 25% to elect instead of 50% are {75%, 100%, 125%, etc., in 25% increment until 2500%}. Any drip chamber made of such material is classified as "heat absorbing" drip chamber, and first we claim the manufacturer (with methods a) b) c) and other methods described above), sell and use of such drip chambers;

Second, we claim "heat absorbing" drip chamber's unique combination (using together) with video monitoring IV controlling devices, including infusion pump, as a particular combined solution for IV monitoring which is accurate, novel, and for the chamber material's sake energy-efficient.

Please also note that, transmission data for our drip chamber slice samples were measured at three different universities (see 056090 and previous applications), and indeed there is room to make more satisfactory and consistent measurement. Therefore, if more authoritative data (such from China SFDA's certified laboratory) is available, they should take precedence over our data, but the relative and absolute absorbance rate increase definition should equally apply (all graphs and tables).

This (allowance of replacement by data from better measurements) also apply to all data, table, graphs in our application since PCT/IB2013/056090.

Light-Proof Drip chamber

There is a particular type of IV set which absorbs significant light energy in the short wavelength region for protecting drugs from high-energy photons. There are different standards and we cite China's national standard GB 18458.3-2005 (^fflfra¾¾S % 3 WA — &ttffi¾¾;¾¾(T¾¾) which requires for light between 290nm and 450nm.

As a result those IV sets look typically brownish or yellow. The intent of light-proof IV sets haven't was to protect drug from high-energy photons from sunlight or room illumination sources (might include UV component, etc.) and in all prior art literatures has never been used in combination with IV video monitoring devices.

The high absorption rate at the defined wavelengths (in different countries we might use WHO or their respective standards and figures) might be used advantageously to increase absorption. In fact, frequently they also absorb significant to above 500nm or even IR (near, mid, far) regions (where lights are "safer" to drug), and the high absorption rate can be used to the advantage for heating. For example, 50% absorption drastically increases heating efficiency some 10 times.

We haven't found prior art teaching the combination of light-proof drip chambers with IV video monitoring devices and in fact it is not necessarily from optical consideration because designers would not choose hazardous (to drug) wavelengths. Because chambers are included in the housing it is not exposed to external light, hence no rationale for light-proof chamber. Hypothetically , one might argue that they use light-proof chambers because nurse might prepare several bags of solutions and administer one after another and the during wait time other solutions are exposed to external light (not in housing). This is not true because:

1) Nurses might prepare several solutions but only one IV set, and the IV set can be removed from on solution bag and attach to other. The practice is not to needle the patient for each IV solution bag. Therefore there is only one IV set for several solution bags.

Therefore in the course of each IV set's use (as a dedicated peripheral to the video monitor, since material, index of refraction, thickness, radius all needs to be controlled) the chamber is always included in the housing. There is no explanation for light-proof treatment/additives other than heating needs.

Hypothetically one might also argue use of light-proof chamber is to make the IV set also usable when exposed to external light (not using video monitoring). Such petty excuses are not worthy of allowance since the prerequisite droplet noise (splash, condensation, abcdS are not even realized so far by prior arts as are solvable by heating, and the argument is clearly based on hindsight.

Therefore we propose the combination of light-proof IV sets with our video monitor. In fact it is included under "heating absorbing chambers", but here we specifically lift the case out to prevent possible equivocations.

Extending to non-PVC material:

Other drip chamber material exist such as TPE (thermoplastic elastomer). For example, German Kraiburg TM9HET and TMOHET are typically used for TPE drip chambers (specially developed for medical applications), and they have been used by manufacturers of TPE IV sets in Shandong province of China, presumably including ANDE Healthcare, a major competitor of Shandong WeiGao group in Weihai city of Shandong which has its own facility for manufacturing TPE plastic. As of July 2014 Ande TPE (no PVC/DHEP) IV set identified by:

1. H¾it^(China National Registration/ Approval No.) 2011.3661327

2. nT(Shandong Province Registration/ Approval No.) 20130179 have been used nationwide in hospitals. We extend the same procedure for calculation absorbance to TPE plastic (for example TM9HET, TMOHET and the above ANDE product), and calculate various heating/light power ratio, temperature rise, energy absorbance directly to TPE drip chambers, and the definition of heat-absorbing chambers, and propose corresponding "heat-absorbing" drip chamber's combination with IV Video monitoring devices. As having been shown in Fig.B.1.2.1-2 and Fig.B.1.2.8-1 and have been discussed in §A.1.2.8 Numerical Simulation, the heater is mounted in front of the viewing window. We have shown incandescent/resistive source as the preferred/optimal embodiment because mid wavelength IR heating is more effective than heating with short wavelength radiators. However, short wavelength heating can still have effect on preventing condensation, and such solutions are practical because there are visible band/near IR emitters with high efficiency. In fact, as having been shown in Fig.B.4-1, PVC material also have appreciable absorption invisible and near IR bands, and in fact it absorbs radiation in all bands. Subtracting reflection, for the cases in Fig.B .4-1 there is still about 4%~5% absorptions in visible and near IR bands, which is enough to prevent condensation with a moderate power radiator/illuminator. And clearly, such a front side radiator/illuminator source is in essence a heater rather than illuminator for the camera, since only a tiny fraction of its radiation could eventually reach the image sensor after been backwardly reflected from the drip chamber. The type of such heater can be LED, incandescent, sodium, fluorescent lamp, metal halide lam, sulfur lamp and in fact any type of energy source, since drip chamber absorbs more or less in all bands.

Sources placed in front of the viewing plane (see Fig.B .11 (3) for illustration) is obviously for heating purposes; sources placed on lateral sides of the drip chamber can in fact only have a small fraction of energy eventually reaching the sensor plane. This is because as shown in Fig.B 2.2-2(2) of the 056090 application, rays need to pass the aperture (or using the entrance pupil formulation) before proceeding to the sensor plane, whereas the aperture size is always limited to increase image quality (viewing angle and spot size are always competing factors; to obtain rectilinear image the lens cannot have large viewing angle as fish-eye lenses). Therefore, lateral sources could in fact have only negligible fraction of energy eventually reaching sensor plane and they are like sources placed in front of the viewing plane, and clearly these lateral sources' purpose is for heating.

We give a clear definition of source direction/location:

Each source have a direction defined by either{the center direction of the beam, or the direction with the highest radiant/luminous intensity }. The vector is then projected to the top-viewing plane as shown in Fig.B.11 (3). The direction/location is defined by the connecting line between drip chamber's center O and the light source.

We define "lateral source" to be those whose direction (as defined above) having an angle with the viewing plane (after all projected to top-view, as shown in Fig.B.11 (3)) between ±45° (and can optionally elect values in the range covered by ±45°, ±40°, ±35°, ±30°, ±25°, ±20°, ±15°, ±10°, ±5°).

And we define "front source" to be sources whose vector as defined above is in front of the viewing plane.

With such definitions the range of "front source" and "lateral source" might partially overlap, however this should not outweigh the fundamental consideration on whether the source is essentially a heating source or camera's illumination source. Therefore what is claimed is both "front source" and "lateral source", regarding of whether the output level is varied between exposure instants or not, and regardless of their type (LED, sodium, incandescent, etc.). Sources from these directions are clearly for heating purposes.

Splash-proof drip chamber Video monitoring Solution

In this and previous applications we discussed comprehensive use on the use of heating to remove condensation. In fact, by projecting sufficiently strong radiation to the drip chamber, splash droplets can also be evaporated. However, drugs have specific temperature limit which if exceeded could cause chemical decomposition or other reaction which might have adverse effects. Therefore, we have to carefully limit the heating temperature to only within the "condensation proof level.

The splash droplets can be prevented from entering viewing window area by combining (a combined solution here) video monitoring devices (condensation prevention mechanism, although highly preferred, in this combination is intended to be optional. We intend a two-fold definition: 1) IV video monitoring device + splash-proof drip chamber, 2 IV video monitoring device with condensation prevention/removal capability /mechanism + splash-proof drip chamber) ) with splash-proof drip chambers. This is illustrated in Fig.B.12(1) and Fig.B.12(2) in which the drip chamber has internal guard (wall or barrel like) to block splashing droplets; various forms are also illustrated in

PCT/CN2013/090173 's Fig.2-1 to Fig.2-6 (incorporated here as Fig.B.12(0)-1 to 6) and text. Fundamentally, splash phenomenon is governed by Reynolds number and the Weber number. Example like Fig.2.4 (Fig.B.12(0) -4) in PCT/CN2013/090173 "intercepts" the falling drips, causing it to break into parts and gradually slide down, thus at its coming into contact with water surface the impact is already small; method in Fig.B .12 doesn't prevent splash at the first place, but blocks their trajectory to the viewing window.

The "prior-intercepting" mechanism can be implemented by needle, oblique cut (V-cut, triangle-cut, etc.), tapering form or others as discussed in PCT/CN2013/090173; "blocking" guards can have rectangular, hexagon and numerous different profile and can have cuts, groove, slits, holes on the walls, and its height and diameter (opening of course should be large enough for drips to fall) can vary in accordance with the geometry of other parts of the drip chamber. To exhaust these combinations is not possible, and the best way is to define "splash-proof drip chamber by effect/functionality :

A splash-proof drip chamber is a drip chamber which during a specified drip rate range, and when the tilting of the drip chamber is less than 1° (or elect values up to until 20°) , splash can never hit upon all or a part(s) of the (front, or front and back) viewing window (of one or several cameras).

Since the faster the drip rate, the more likely splash would be resulted, the speed range is usually specified by [low bound, higher bound], for example, lower bound might be chosen from {0, 10, 20, 30, 40} and upper bounds might be taken from {60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, .until 300} (there are indeed some very fast infusion like rapid/pressured transfusion to operation room patients). The viewing window (or its part(s)) is usually specified by a rectangular area such as (W^XH) [4.571, 4.571](2.2854mm is the radius of spherical 20drips/ml standard drip), [6, 6], [8, 8], [10, 10], [12,12], [10, 12], [10, 15], [12, 15], [14, 16], [15, 20], [20, 20], [20, 25], [20, 30], [30, 25], [30, 35], [30, 40], [35, 40] etc. and can elect any therefrom.

Note: These window size changes, associated power, temperature and other values should change correspondingly. The procedure has been illustrated in many places in this application, for example, Table.B-2's.

To our best knowledge, in patent literature there has never been any drip chamber specifically designed to prevent splash to any of the "viewing area/window", either with respect to the human eye or machine. US Patent 4465479 taught "truncated conical member" form splash guard at drip chamber top, but not as our "wall" below the viewing area.

Please also note that although we explicitly draw a lens (essentially a light-directing device to improving heating efficiency, heating can also be achieved from back illuminator/simultaneous heater), or lateral, or higher or lower sources. All analysis in this application are based on energy and thermodynamic figures, and apply to light/heater sources from all directions.

The optical monitoring solution (a technique + apparatus combination) only matured after overcoming the following four obstacles:

1) Condensation (must use heating; hours of mere condensation often results in surface droplets larger than shown in Fig.B.(l) ), and its validity was established only after the large amount of calculations in 056090 and ensuing applications.

Note: in addition to heating, in application US 13897578 and US 13903924 we have also described cooling the drip chamber, (or tube above the chamber, see ^Cooling and combine Heating/Cooling), which is also capable of prevent condensation. Hence, the splash-proof drip chamber can be combined with

2) Splash (must use splash-proof drip chamber): its adverse effect to video monitoring is the same as condensation; however, without proving condensation can be prevented first (proof by show chamber surface heats to higher temperature than saturated vapor, using HITRAN gas spectroscopy data), merely preventing/block splash still could not result in clear image. The two approaches must be taken at the same time.

To our best knowledge and effect, all prior arts, if they ever addressed condensation/splash problem, all suggested treating them using filtering and various image processing techniques.

Failure of prior art:

"Drip counter", particularly those using infrared emitter/transmitter (in essence also a machine-vision application), existed for more than a decade and but has never become a widely -adopted/reliable device. Aside from environmental light interference, it is mostly affected by condensation/splash droplets and tilting of falling trajectory (discussed below). These problems remain unsolved for at least over a decade.

In fact, there has been significant financial investments to overcome limitations of existing infusion pumps. For example, there has been over $40M investment by a US company over a decade, yet product is still in improvement.

3) Sensitivity of volume measurement to drip chamber radius, thickness variation, and to droplet's distance from the camera. These are answered by our comprehensive calculation in IB2013/056090, before which no literature have ever shown and compared the strength of these factors, and whether accurate video monitoring of droplet volume is viable (Followers acts only after those in the leading positions has proved thing's feasibility!).

4) Tilting of droplets. This is a problem existed for over a decade from IR drip counters. Tilting change distance microscopically for each pixel's distance and magnification to the camera, and must be calculated and corrected using methods in part C of this application.

The combination is like marriage producing a brilliant child, yet the match is only discovered after recognizing and solving the perennial condensation problem, but in theory and in practice. We regard this as one of our most important contributions.

Hence, what are claiming is the combined solution, including monitoring camera(s), splash-proof drip chamber, heating sources or lights with sufficient heating ability, and tilt/droplet distance measuring/calculation devices (might be optional if the device never tilt, like for a heavy infusion pump place on the table).

In practice, optical IV monitoring require precision-made IV sets (see D, R, T variation calculation in 056090 application), therefore IV sets are not arbitrarily chose, but are made specifically for video monitors. Therefore, these special IV sets should be viewed as "special/dedicated accessories" for the monitor, like ink cartridge for each particular type of printer, whose parameters are specified by the monitor manufacturer. Therefore, a combined solution is formed with them and the optical measurement device, rather than separate, arbitration combinations.

Contents of PCT/CN2013/090173

To comply with PCT and national law requirements the key elements of priority application PCT/CN2013/090173 is physically incorporated below, with its drawings incorporated as Fig.B.12(0)-l to 6. The 090173 application had already completely and unambiguously suggested the combination of IV video monitor and anti-splash / splash-proof drip chambers which eliminate splash droplet at the first place instead of dealing it with later image processing.

Its "background" says: "The purpose of this invention is to create a device that could achieve accurate volumetric measurement of falling droplets during an IV administration, and it works by video and image processing techniques. Splash droplets on drip chamber surface create undesirable noises in images. Whereas conventional idea is to remove these slowly changing "low frequency" "background noises" with filtering techniques like image subtraction, we show below that easy modification to drip chambers could eliminate them completely.

The closest prior art we found was US patent 4465479 Air Vent Splash guard for Drip Chamber, by Charles E. Meisch."

Its "Summary" says:

"This application discloses novel drip chamber designs that could effectively eliminate splash droplets. We show that needles and extended tube inside the drip chamber could "cleave" falling droplets and absorb their impact before droplets interact with the liquid surface, thus preventing splashing droplets."

Its "Specification" says:

"It is always preferable to get clear image in the first place than to remedy imperfect images in the second place. Therefore we present designs of drip chambers in which no splash droplets could form.

In ordinary drip chambers such as in Fig 1-4 (see Fig.B.12(0)-4), splash and condensation droplets could severely obscure the droplet image to the extent that no image processing algorithm could successfully recover the true image of the droplet.

This application deals with splash droplets alone. We have observed that

1. splashing happens only when droplets falls into relatively "deep" liquid.

2. when the drip chamber is tilted an angle so that falling droplets hits lateral surfaces directly, there is no splashing droplets.

These observations suggest that in order to prevent splashing, we could use solid objects to "intercept" the droplets before they fall into and interact with the lower liquid.

There are two types of objects that could intercept the falling droplets:

(1) Tube (could maintain their shape, but generally soft) extended up from the lower of the drip chamber

(2) Obj ects harder than tube .

And all the designs are shown in Fig 2-l(see Fig.B.12(0)-l)."

It has also given several design examples and experiments: "Example (Needle)

A very typical example is shown in Fig 2-4(see Fig.B .12(0)-4) where there is a tapered needle placed at the lower side of the drip chamber, and that needle is connected with the tube below.

The tube is tapered so that it will very effectively "cleave" the falling droplets when the droplet touches it at its central point. The tapered structure also serves to spread the "clashing shock/impact" over the entire tapered structure from higher to lower when the "pierced" droplet "slides" gradually down the tapering structure. Therefore when it reaches the liquid surface its momentum has already been distributed evenly into 360° direction and there will be such a "soft landing" where the impact is too small to create any splash droplets.

On the lateral side of the tube there are holes (one or multiple). This is because is the lower end of the needle is connected seamlessly with the lower tube, the liquid cannot pass into the tube. We do not exclude the case where the holes are at the top of the needle, and in fact we have tested the cases with holes at the top of the needle and these needles could effectively prevent splashing while passing liquid down to the tube (see Fig 2- 3(see Fig.B.12(0)-3) for shape example and Fig 2-6(see Fig.B.12(0) -6) for photo) . However, with top holes the liquid surface will eventually reach to the same height as the needle's top, and that point the needle's ability in preventing splash droplets will be reduced. Therefore, Fig 2-4(see Fig.B.12(0)-4) shows actually a preferred embodiment.

Deviation in shape

Deviation and modification from Fig 2-4(see Fig.B.12(0)-4)'s preferred embodiment include;

1. Fig 2-3(see Fig.B.12(0)-3), where the tapering structure is replaced by normal round/circular structure, and that the needle might be hollow and/or with hole on the top side.

2. Fig 2-5(see Fig.B .12(0)-5), where there can be multiple tapered (or not tapered) needles

(fork/trident-like, etc.) to further break falling droplets into even smaller parts can further reduce impact shock.

The material of these needles can be any relatively "hard" material and there is really no restriction as long as the material is safe to drugs and its rigidity satisfies the requirements. They may be metal, plastic, rubber and all types of suitable materials.

Tube as needle

In fact tubes can also serve the same function. From the lower side of the drip chamber if we insert the tube a bit further inside (in contrast to "just-in" as found in ordinary drip chambers), they will then look behave like needles. We suggest two modifications:

1. Piercing lateral hole(s), so that the liquid level will always be the same as the hole(s) and the top of the tube can stand above the liquid surface. The part standing above the liquid level surface can effectively "cleave" falling droplets based on the same principles described above. See Fig 2-2(see Fig.B.12(0)-2) for example.

2. Cut the top head of into an acute shape (like a cut bamboo), as shown in Fig 2-2(see Fig.B.12(0)-2). The key is that the circumference does not have a uniform height. The liquid surface in the drip chamber will always be the same as the lowest height on the circumference, and those lengths of the circumference higher than the liquid surface will serve to "cleave" the falling droplets. We give several definite (as the patent law requires) numeric angles for the cut, starting from 5° (0° is flat, corresponding to ordinary flat cuts), with 5° increment up to 85°.

Other more sophisticated cuts like the "V-cut" shown in Fig 2-2(see Fig.B.12(0)-2) can also be used which is based on the same idea above and they should be treated as equivalent.

All the shapes/materials we described are summarized in Fig 2-l(see Fig.B.12(0)-l).

Experiments

We have tested the solutions in by fixing sections of short tubes and needle vertically into a container (filled with liquid), surround them with drip chamber surfaces, and using an IV set to let IV flow fall on to each respectively, and verified that the solutions eliminates splashing effectively ."

Its "abstract" says:

"This application discloses novel drip chamber designs that could effectively eliminate splash droplets. We show that needles and extended tube inside the drip chamber could "cleave" falling droplets and absorb their impact before droplets interact with the liquid surface, thus preventing splashing droplets."

Both the title of PCT/CN2013/090173 "Anti-splash drip chambers" and its contents, when read as a whole, clearly suggested that we intended to teach the combination of video IV monitor and all types of drip chamber which could functionally/effectively prevent splash droplets. This current application adds more implementations into its scope.

On the variations of splash-proof drip chambers

The general functional definition (from effect perspective) on what is "splash-proof drip chamber has been given, and two mechanisms ("intercept/cleave/speed reduction" and "block splash") has been exemplified. Here we give more examples:

The methods of splash prevention can actually be classified analogously like the schemes of noise reduction:

Method (a): The drip chamber might have a wider lower part than top as in examples in Fig.B.12(3). The pressure of air in the chamber approximately equals the atmosphere pressure plus the pressure caused by the liquid height from IV solution bag to drip chamber, hence the air need to be compressed, so liquid always rise to a certain level within the drip chamber. A wider bottom makes this rise small (consider the volume ratio) so the liquid surface is low. Although low surface imply that falling droplet obtains more kinetic energy when falling onto it, when the distance is large enough splash droplets would not be able to reach the height of the viewing window. Two examples in Fig.B.12(3) has tall "narrow" part of drip chamber which also limits the splash droplets' direction and servers to "block" them. With properly adjusted geometry (splashing study is also the capability of many CFD software) designs with a wider lower part can be guaranteed to completely prevent splashing onto the viewing window. The tapering from between "fat" and "narrow" part in Fig.B.12(3) is primarily for aesthetics/filleting(aesthetics & manufacturing considerations), and in fact the embodiment can take numerous shapes.

The left of Figl2.(4) put a "web" above (or below, but not too deep) the liquid level, reminiscent of steel web filters of some teapot. When droplets the web "cleave" them apart and reduce their momentum, and if there are very small splash droplet coming up they again block them. Their mechanism is similar to CN2013/090173's tapering tube or needle cuts.

The right of Fig.B.12(4) is asymmetric in that droplet intentionally being directed to either farther or nearer from the viewing window. The chamber surface side droplet close to might "smooth" the falling and block the splash droplet's rising at an early stage (like the rising-stage interception of missile defense system), and because of the asymmetry splash droplet will not have enough energy to reaching the other side which is typically the viewing window side. A proper design of such can prevent splash droplet for reaching both the front viewing window (near camera) and the back viewing window (near the back lighting source).

The ultra-fat drip chamber in Fig.B.12(5) increases distance from location of droplet's impinging the liquid surface to the viewing window (front and back) so no splash droplet would reach viewing window. The right image's slim lower part serves to block, and the fat middle part increase the distance.

Fig.B .12(6) left image added a round of "blocker lid" beneath the viewing window, which can also be partial instead of full circle as in the right image.

Apparently it is difficult to exhaust splash-proof designs as implements can always be mixtures of the types/principles, or some new designs. However, the specific combination of splash-proof with video monitoring IV system is unknown in prior arts, and their use in combination with heating mechanism solves the image quality problem in IV monitoring once and forever. We claim the unique integrated solution (electro-optical-mechanical device plus viewing window splash-proof drip chamber) and the drip chambers are regarded as accessory /peripheral of the electro-optical- mechanical system. Other reasons for droplets on viewing window (and the backside of the window) (exclude condensation)

Include back viewing window :

We explicitly state here that design, effect and treatment (heating, etc.) for the front window apply equally to the back window. If droplets are on the front window they obscure and change the image, and the restored image at most times loses some accuracy. If they are on the back they can usually be viewed as background and image processing is relatively easier. However if they are too many back window images, they might also cause problems to the algorithm. Therefore, all apparatus, configurations, combinations and methods described in this and past (chronically from our PCT7IB2013/056090) applications are all applicable to back viewing window.

Splashing happens after falling droplet hits liquid/solid surface and is not the only reason for droplets on viewing window. We describe another four causes of droplets on the viewing window: a) When chamber is tilted, liquid can reach window from lower and leave a water layer or droplets.

b) When chamber is tilted, droplets from dripping mouth could fall onto water

c) The drip chamber might be occasionally sloshed/shaked/tapped (alone or along with IV monitoring device), or subject to acceleration during movement.

d) During priming there might be high flow rate for short duration.

The designs discussed in §On the variations of splash-proof drip chambers and shown in Fig.B.12 are extended and summarized in Table.B.4-1 to Table.B.4-3.

In Table.B 4-1(1) we divide drip chamber into {upper, middle(optional), lower} parts and for each part we give design examples on how {a,b,c,d} causes can be handled. We see that roles are overlapping: the design to achieve one purpose might incidentally achieve others simultaneously so it is necessarily to enumerate all or as complete as possible of them for considerations of legal formalism.

Table.B 4-1(2) shows that high liquid surface can also be used to lower Reynolds/Weber number. Optional guard can handle causes {acd}.

Table.B 4-2 shows designs to break droplets and/or slowing them down to prevent splash. Both Reynolds number and Weber number [.

Note: the tables typically show half profile on one side of the vertical. It is not necessarily symmetric and the profiles along different cut planes could differ. The design intent is only to prevent droplets (exclude condensation) from forming on the window. Italic S in annotations denotes splash.

Intercept, cleave, buffer: break

droplets into smaller; buffer,

absorb impact energy;

Slope (90° in the 5^th is an

The position of these

extreme of slope) gradually

1. Interceptor/cleaver

slows speed down so impinge on

2. De-accelerator

surface then minimal;

3. Buffer

2^nd correspond to Fig.2-5 in

are positioned such that their tip stand above the liquid surface, or CN2013/090173 (see

their tip be no more than [2 times the height of the droplet(allow Fig.B.12(0)-2 to 5), representing

electing 1.9, to 0.1 in 0.1 decrement, or absolute value multiple tapered shapes;

2.28539x2x{2.0, 1.9, 0.1}, 2.23539 mm being radius of spherical 0.05mL droplet)] below the surface. Even if the tip of them are below the surface, they can still absorb significant power of the impact hence reduce or eliminate splash.

1. The material of {interceptor/cleaver, de-accelerator, buffer} include rigid as well as more flexible material. For better buffering/energy absorbing effect, they might use

• Porous

Table.B 4-3 : the cross-section (at any height) profile might assume various shapes. Left: ellipse with different e; Middle: polygon, trapezoid, parallelogram; Right: arc + straight lines, petal (arcs).

If viewing window (front and/or back) are flat then the optics is relative easier; for cases in left of Table.B .4-3 with e≠ 1 optical correction in 056090 application also apply (either analytical calculation or computer simulation/calculation); correction of 056090 also extend to right cases and in fact all cross-section shapes.

The shapes cannot be exhausted but the basic methods in part A and part C extends naturally to all.

All these cross-section shapes are realizable by such as material extrusion, and possible then assemble from different parts.

Synthesizing→ clear image drip chamber when condensation is not considered From the above definition we see that it might be improper to simply call these designs as effect X-proof droplets since the simultaneous of several or all might causes ambiguity in legal delineation and one effect might be construed as the simultaneous incidental effect(s) of another(s).

We give formal definition of a/b/c/d-free drip chamber respectively in which we used letter to indicate the effect: a-f ree: drip chamber which when tilted to between 1 ° and 80° (lower bound and upper both allow election in 2° steps in range [1°, 80°] if overlapping with prior arts), the liquid surface does not reach the viewing window (front or back). b-f ree: drip chamber which when tilted to between 1 ° and 80° (lower bound and upper both allow election in 2° steps in range [1°, 80°] if overlapping with prior arts), the liquid surface does not reach the viewing window (front or back). c-free:

The exit speed of droplet from chamber drip mouth depends on dripping rate, viscosity, pressure and material and for simplicity we assume 0 m/s when just breaking. Take falling distance as 15mm corresponding to the 15 xlOmm window we used extensively in calculation.

For simplicity assume vertical acceleration is zero, denote horizontal acceleration as a.

Note that in Fig.B.12(5) we might have chamber particularly wide at windows area height, and in practice we also see chamber varying in diameter, and the calculation above made assumption on the acceleration direction. To accommodate these variations we propose the definition of c-free drip chamber as drip chamber in which when subjecting to an average horizontal acceleration between [0.2m/s, 20m/s] (allow election from 0.2m/s to 20m/s in 0.5m/s step for both upper and lower bounds) :

1) Droplets falling from dripping mouth doesn't not hit viewing windows directly

2) Lower liquid in the drip chamber does no rise, tilt (due to acceleration), from turbulence or for any reason reach the viewing window (front or back) . d-free: drip chambers in which during the continuous flow condition at priming (typically when clamp is released) there can be no droplet hitting the viewing window(s) .

S-free/ roof: already defined in §Splash-proof drip chamber + Video monitoring Solution.

And a clear image / noise free drip chamber is defined as drip chamber either designed from any or more of the principles discussed using principles in 1. §Splash-proof drip chamber + Video monitoring Solution

2. §On the variations of splash-proof drip chambers

3. §Other reasons for droplets on viewing window (and the backside of the window) (exclude condensation) or achieved any or more of a/b/c/d/S-fiee effect. The requirement doesn't not require that all a/b/c/d/S-ΐκ be achieved simultaneous since not all are of equal importance.

In conclusion, a clear image / noise free drip chamber (excluding condensation consideration) is the comprehensive definition subsuming other formulations and the subsumed alternative formulations have overlapping or equivalences so they need to be formulated from different perspective separately.

We firmly believe that these alternative formulations are necessary in a patent literature which as a legal instrument shares much of the same formulism characteristics as of mathematics.

Proposed shapes

We have proposed drip chamber bulk part (not including dripping mouth, tubing, fillet, etc.) shapes with exact dimensions shown in in Fig.B.25-1 to Fig.B.25-6, all them satisfy the requirement above and could prevent S and part or all of {a,b,c,d}, . Unannotated thicknesses defaults to 0.5mm, varies between 0.3 and 2.5mm. The cross-sections allow changing shapes according to including Table.B 4-3 where we can equate the radius (Fig.25-2 to Fig.25-6 are all ortho-projection and unannotated dimensions can all be measured in images ) in Fig.B.25-1 to the major (longest) or minor (shortest) or any meaningful cross-side measurements in Table.B 4-3.

We further address two considerations on the drip chamber design which is consequential of its use in video monitoring. Traditionally, the surface of liquid chamber cannot be too low for two reasons:

1) If the surface is too low, the Weber number of droplet it high upon hitting the surface and due to the impact the small bubbles formed at the collision can enter the liquid. The faster the dripping rate and the longer the falling height, the more bubbles accumulates. During the course of administration the bubbles becomes more and more and could even merge into macro bubbles, which under some circumstances would even enter the body through the tube. This is well known in IV administration and is one of the major safety concerns. Therefore, the standard practice (for most hospitals in China) is to require the height of the surface to be 2/3 of the height of the drip chamber, and stipulates that the surface height should not be lower than ½ of the chamber. The absolute dimension for the vast majority of drip chambers are shown in Fig.B.26-1 and Fig.B.26-2. [one reference f|»«j«Sⁱ$lJ¾*¾-P/rl§;S¾¾¾ail - ¾ rUA RE B¾]

2) When solution contains organic molecules / large molecules / cells, etc., the impact could have effect on the solute/drug. There has been studies shown that during transfusion, the liquid height variance induces statistically significant variance on the red cell count and other metrics [ ((ί^3 ¾ί> » 2001 09 M fit JftLBi¾^ RftlWftffiitSJt^tEftaJltm-ffiftlW^]. Therefore it is necessary to limit the surface to above a certain level. However, during IV video monitoring we have to a) Maintain a viewing window large enough. b) The actual image processing area is below the dripping mouth.

IV sets might have different nominal droplet size (20 drops/ml→ D=4.47mm, 30drops/ml→D=3.992mm,

15drops/ml→D=5.03ml) and we at least keep window several times larger than droplet dimension D, and for the splashing droplets, even if they don't hit viewing window, we need to make sure that the agitated oscillating surface and splashed droplet are far/low enough from the measuring area, hence there is always a tendency to make the surface lower.

Without using video monitoring the 1) and 2) requirement can be satisfied be increasing droplet height; with video monitor the two competing requirement necessitates lowering the surface. In order to satisfy requirement 1) reduce bubble and 2) reduce impact on drug/solute, we propose adding a) Interceptor

b) Cleaver

c) Buffer

d) Web/string/net

As shown in Table.B 4-2 and CN2013/090173, allowing materials to be

Porous

Fibrous

Sponge(ft¾l, )-like

Foam(ifi¾)-like

Rubber-like (including Thermoplastic elastomer)

As for the location of them, we stipulate that the top (tip) of {interceptor/cleaver/buffer} height of web/string/net be at least lower than 1.2 times (allowing electing from increment in 0.2 times if conflicting with prior arts) the nominal droplet spherical diameter of the IV set, yet if beneath the surface, the tip/web height be no more than 2.0 times (of nominal droplet spherical diameter) lower than the surface (allowing electing between 2.0 and 0.1 times below the surface in 0.1 increment).

The requirements of 1) and 2) are medically necessarily. In infusion pumps there are bubble monitors to detect the presence of 1), therefore our reducing them using {interceptor/cleaver/buffer/web} is duly required when the combination with video monitor necessitates lower of the surface.

As an incidental effect, the inclusion of {interceptor/cleaver/buffer/web} might eliminate be reduced splashing droplets due to lowered impact (due to sliding, friction, split / surface-area-increased droplet). However, the motivation of 1) and 2) itself is scientific, strong and convincing enough to justify its primary use, rather than secondary or incidental effect.

For the purpose of either reducing collision velocity (Weber number), or mitigate blood cell collision, or prevent formation of droplet, we might also reduce the distance between falling mouth and the standard (according to IV set's recommended use practice) droplet surface. We propose using drip chamber which keeps this height/distance 6.0 (allowing electing in 0.1 decrement down to 1.2) timer smaller than the nominal spherical diameter of the droplet.

Small splash droplet surfaces: Chemical and Geometric Solutions

Reference:

[1] [The effect of polymer surface on the wetting and adhesion of liquid systems - J. Adhesion Sci. Technol., Vol. 21, No. 3-4, pp. 227-241 (2007)]

(2) [H. Murase, K. Nanishi, H. Kogure, T. Fujibayashi, K. Tamura and H. Haruta, J. Appl. Polym. Sci. 54, 2051 (1994).]

(3) [H. Murase and T. Fujibayashi, Prog. Org. Coatings 31, 97 (1997).]

[4] [E. Pierce - Understanding of sliding and contact angle results in tilted plate experiments, Colloids and Surfaces 2007]

In cases when splash droplets do reach droplet, if they can be controlled small enough, image processing algorithm will be able to handle them. As having been discussed in the three cited articles above, the droplet side on a 90° surface is characterized by volume-sliding angle measurement, not merely tension and contact angle. As having been shown in Fig.4 of [1], silicone (safe and has been used for IV tubing) has the smallest sliding (among [1] material, not absolutely among all materials) angle for fixed volume and the paper also shows that at 90° the largest droplet volume is about 3uL. PTFE, though having lowest friction coefficient and highest contact angle, actually have larger contact angle.

However, the experiment in [1] not covered case when sliding angle is larger than 90° (somehow pendant). At 90° or smaller the polarity /hydrophobia property of PTFE might repel the sliding down property of droplets, but not necessarily when angle > 90°. This is intuitive plausible since inclining angle >90° gets closer to the case of pendant droplets for which case, highly hydrophobia material tends to only hang small droplets. When angle > 90° (practically even can be >90°+45° or any as we can optically correct image, or shoot from an downward-looking "oblique angle"), the low surface tension with PTFE and other hydrophobia surface might result in only very small droplets.

It is therefore propose two ways to minimize splashed droplet size:

We define any "smaller" below to be below the measurements of TPE and PVC samples, either or both. By using material that is more "sliding"

We propose reducing surface tension between liquid and inner surface. By chemically alternating material composition this is achievable. First, Polytetrafluoroethylene (PTFE, Teflon) is medically and has been used in intravenous catheters for decades, and is plausible suitable for using with drip chambers.

Second, PTFE has nominal coefficient of friction (we are not limiting to surface tension only) that is 0.05, far lower than the typical 0.2 to 0.3 value of PVC. TPE exhibit rubber-like property and has much higher friction (many types above 0.6).

Therefore we propose treating PVC, TPE or any chamber material, by additive or surface coating, so that

• Nmerically the COF is more than 0.02 (allow electing 0.03, up to 0.3 in 0.01 increment) lower than typical PVC (can be experimentally defined by the four samples used in our absorption measurements), or typical TPE (experimentally measure the two Kraisburg sample).

• Relatively the COF more than 5% smaller than the same said above sources, allowing electing in 1%

increment up until the ratio between TPFE and the material.

For surface tension, water-PVC is typically 37.9 unit and contact angle 84.5° where water-PTFE has 19.4 unit and 109.2° angle. Hence we define

• Numerically: using any material X whose water-X surface tension is more than 0.5 unit (allow electing

0.6, ... , 0.7 in 0.1 increment all the way up) smaller than that of its untreated state (typical medical tube PVC and the two Kraisburg product can experimentally define PVC and TPE, TM0HET AND TM9HET).

• Relatively, contact angle be more than 3% larger than its untreated(un-libricated) state.

We propose the use those low COF and low water-surface tension material for drip chamber. It is particularly recommended these materials be used to for viewing windows that are more than 90° steep (somehow pendant). We allow choosing from any angle below 90° up to 175° since optically our sophisticated methods discussed in 056090 and later applications allows correcting all distortions (electing reduced ranges in ±5° when overlapping with prior arts).

We also propose materials characterized with low sliding angles. This is defined as performing the measurements as defined by [1] (either using the mentioned surface preparing or not), where we allow for controlled drop volume for any value blow 10uL (preferably < 3uL), measure the sliding angle. If that is more than 5% smaller than {TPE, PTC, optionally include other less common chamber material}, we define the material as being "sliding material".

Specifically, we propose using existing material (PVC, TPE and other less common types) with silicone (mixing or surface coating). If the material has the said measured sliding angle in 20% closeness to silicone or even smaller than that, it is also a highly preferable "sliding chamber material". An alternative definition of sliding angle is according to [4] in which it is either a) measured when drop placed on inclined surface;

b) measured when droplet placed on level surface, then inclined;

We stipulate that either a) orb) may be used to judge whether the material is our "sliding material".

One top-view drawing is shown where the droplet suspends on chamber surface. The top-view cut profile of the droplet which are somehow sessile and "flattened". If we increase radius (for example, from the standard 7mm nominal used in thermal simulation models in this series of applications to 9mm), then if droplet size if fixed, contact circumference will decrease hence it will fall. Therefore with increases inner surface radius we will have smaller radius. The middle shows radius approaches a flat plane (infinity), and the right drawing shows it even become "negative" (optical terminology) in which even our daily experience would convince us that droplet size will become smaller. Smaller droplets are easier to process by image processing.

In definition, we use the two radii rl and r2 definitions in ^Facilitating cooling. If at one or more of the cut planes at the viewing window area (not cooling area) the radius of curvature is larger than the numeric values larger than rl or r2, become infinity, or the radius of curvature is even negative (local center outside chamber), then it is also defined as a "tension- wise flatter-or-convex surface".

And we propose the use of "tension-wise flatter-or-convex surface" to reduce suspending droplet sizes (can due both to splash and condensation).

Note that we also proposed the >90° oblique shape above for facilitating droplet falling/sliding.

Hardened Support

For "soft" material like silicone, we propose making the "bone'Vsupport of drip chamber with more rigid material to keep a rigid geometric/optical shape, like rib or cage. The soft material attaches to the harder material. Therefore, the combination is characterized by a "soft area-hard bone" combination. Calculation without fixed Heat Transfer Coefficient

The previous calculation used a fixed heat transfer coefficient (A, use acronym HTC below) was assumed to be a fixed value 2.0 W- m^~2 K^_1, this is based on Table 5-1 "Typical values of convection heat transfer coefficients" in [Cengel - Heat Transfer: A Practical Approach] . We also referenced default heat transfer coefficient value in Solidworks Flow Simulation and Fluent software examples, which is typically between 5 and 6 W- m^~2 K^_1 for metal enclosures and when natural convection (not forced) is considered. The nature of the HTC definition however, are mostly computational oversimplifications when differential equations needed to be solved by manual calculation, and much of the handbook values were empirically determined. HTC also differs with ΔΤ, and assuming a uniform HTC over an area often lead to solutions with temperature more uniform than it would be in reality.

Nevertheless please note that HTC is a number very difficult to determine. One method calculates Nusselt number first then solve h in terms of that, but obtaining Nusselt number one has to in turn calculate Reynolds number and Prandtl number, implying that the flow dynamics must be first solved.

Alternatively, instead of using fixed HTC we also conduct another round of simulation letting the Solidworks Flow Simulation dynamically calculate the convective heat transfer (and the implied HTC) on the fly. The finite volume method used by Solidworks Flow Simulation is well-suited for natural convection driven flows and is known to have robust convergence as it solves for the conservation equations locally at every cell. The alternative results for §5.Numerical Range and Restriction on absolute drip chamber surface temperature rise are respectively provided below, the final choice (as for which dataset used) should be determined solely by the yardstick of their agreements with reality.

150 1.00000 683000 42.76% 340.11

200 1.33333 910667 57.01% 351.65

Table.B-2(2) (Dynamic heat transfer coefficient)

^{1 1} including the "viewing plane luminous flux" definition; The calculation for this column is done assuming heating power coming from incandescent source, and infer the maximum illumination power (not limited to visible spectrum but we used Fig.B.9-3 's photopic values for illustration). The lux in the 3^rd column divides 6.69082 in Fig.B.9-3 to get photopic power, then again divide 238732 lux from Table.B-1 which is a practical and realistic illuminance according to Table.B-1.

For example, as in § J. Numerical Range, we can use ratio threshold values above, like 0.29%, 1.71%, 3.42% 5.7%, 11.4%, 22.8%, 42.76% or any other value in the table, or within range of the table, to define "heater" classification thresholds.

Table.B-3

Fig.B.15 also includes results for thickness = 1.0 and 1.5 cases, and theirvalues can be plugged in the same manner into the above table for fixed heat transfer coefficient cases. Priority claim:

Contact heaters were first introduced in applications US 13897578 and US 13903924, both filed in May 2013. The contents below discuss cooling of liquid, and we claim the above priority for those although described in the two earlier applications.

Contact heaters were later discussed with CFD simulations in the 056090 applications, which we also claim priority to. The corresponding figures were Fig 1.1-1 to Fig 1.1.1-2 (see Fig.A.1.1-1 to Fig.A.1.1.1-2).

Fig.B 16-19 illustrates different constructions of the heaters (can also be used for cooling after some alterations, please refer to section Cooling and combine Heating/Cooling).

Fig.B 17-1 shows an example of the front heater.

Fig.B 17-2.1 and Fig.B 3-2.2 show an example of the front heater composed of two movable parts Fig.B 17-3 shows a "patch" like front heater at the top. Fig.B 17-4 shows a ring-like heater.

Fig.B 17-5.1 and Fig.B 3-5.2 show different views of aback heater.

All the heating methods described so far have cooling as their duals (opposite/complement). This is because our ultimate goal is to keep temperatures of specific areas of the drip chamber's inner surface above dew point, to achieve these we can either increase the temperature of these inner surface areas, or lower the dew point. The dew point can be lowered by lowering the humid air temperature.

Therefore if instead of explicitly transferring heat to those drip chamber surface areas, we may remove heat from the air, or from the liquid which will in turn lead to the lowering of the temperature of the air, and we somehow maintain (or raise, or lower it but keep it still higher than air/liquid) temperature of specific areas of the inner surface of the drip chamber, then they will still be above the dew point.

The wrapper arrangement in Fig.B. 21 looks exactly like a dual (opposite/complement) of Fig.B. 16-1. Whereas in Fig.B. 16-1 heat is explicitly transferred to windowed area and back surface area, in Fig.B. 21 heat is removed from the two side surfaces which could lower the air temperature inside the chamber. For the drip chamber surface area facing the camera, as well as the it back (far) side, because the heat conduction of the drip chamber's surface is slow, it is possible that the temperature of these areas decrease slower than the air inside the chamber, and in this way we have successfully keep these areas' temperature above the dew point. If we would like to cool the liquid, we could move the patch-like structure in Fig.B . 3 -3 to the bottom to make it vertically in contact with the liquid (separated by drip chamber material) and it would effectively lower the liquid temperature down, and consequently temperature of the air above the liquid.

Location and length of cooler:

1. First, we may cool area of the drip chamber above the liquid surface. The purpose is to make viewing window temperature (inner surface) > air in contact with it, so we may use cooler to cool other parts, for example lateral part(s), and when these area become colder it will cause air distribution (there will be uneven temperature/density distribution and convection), and humid will first condense on the cooler areas, but not on viewing window.

Simulations for heaters have already been included and for cooling is essentially the same. Because energy exchange does not involve internal gas directly, the effect on gas is only secondary, hence we are guaranteed that at cooling locations the drip chamber's area is always lower than the internal air, and this location(s) will be the only site(s) of condensation.

2. Second, we may cool including below the liquid surface. As we have talked above, the patch or whatever heater can be moved to lower location, hence the cooling is first on liquid, then liquid cools the gas.

Of course since liquid surface level can vary, the cooler may heat both 1 and 2 above simultaneously.

3. Third, we may place cooler at sites on upper, top or above the drip chamber. For example, cooler (cooler location defined by distance between bottom of the cooler and top of the drip chamber, allow election from 1 μιη, 0.1mm, 0.2mm, 0.5mm, 1cm, 2cm above, ... values in 0.01mm increment until 180cm) may surround directly the tube at these locations, and because they are in closer contact with the liquid (if we put them in contact with lower part of the drip chamber, we cannot press too firmly lest it will alter the shape of the chamber to create optical distortion, but no such concern when cooling the tube directly above) we can have higher efficiency.

We may further increase length of the cooler, as what have been done for some liquid warmers, so that liquid is cooled over a larger distance rather than at a single site in order to cool them sufficiently.

We define cooling length for cooler cooling the tube directly above chamber to be the aggregate length over all cooling location above the drip chamber, and the length is defined to be between 1 micron (lus; in practice, 0.1mm above is more realizable) and 180cm. The lower limit of the cooling length is allowed to elect from{ Ιμβ, 0.1mm, 0.5mm, in lmm and above in 0.01mm increment to 100cm}, the upper limit of the cooling length is allowed to elect from {2μβ, 0.1mm, 0.5mm, and above in 0.01mm increment to 180cm}, and stipulating that the upper limit is always larger than lower limit.

Novelty: To our best knowledge we so far have not find any cooling device so far in the widest range of IV administration/care devices with extensive searches, not to mention their combination with video IV monitoring devices. Because the dual (opposite/complementary) relationship between cooling and heating, all the heating methods/sources listed in Fig.B. 20-1,2 and/or discussed above, as well as all arrangements from Fig.B. 16 to Fig.B. 21, and the entire set of heating apparatus/methods in applications filed temporarily between US 13897578 and the current application ,can be used on cooling after straightforward modifications.

It is also obvious that cooling/heating can be applied simultaneously to create the relative difference so that the temperature of specific areas of the inner surface of the drip chamber is above the dew point.

The essential contents of priority applications US13897578 and US13903924 are incorporated here to comply with PCT and International Laws:

[START: US 13897578 and US 13903924 contents. Please note they are excerpts and doesn't constitute "headings" or divide the current application ]

Fig 2 (Fig.A.1.1-1) shows front and back drip chamber surface heaters.

Fig.B .16-1 shows a camera facing the front face of the drip chamber emphasizing the front heater. The front heater has an open window to allow camera to see the drip chamber. Part of the back heater can also be seen. Only five faces of the housing is shown, but it is intended to represent the whole enclosing housing structure.

The "wrapper" -like structure are heating elements and the shape in the drawing is only for illustrational purpose. The pair of small blocks represent transmitter (the thinner and longer one) and the receiver (the flatter one), but they can also be single-element sensor such as a camera, in which case it needs to be placed farther at an appropriate distance from the drip chamber to get clear images. For non-imaging detectors (transmitter-receiver / sensor) the distance also depends on the specific elements, although the requirements tend to be looser.

Fig 3 illustrates different constructions of the heaters (can also be used for cooling after some alterations, please refer to section Cooling and combine Heating/Cooling).

Fig 3-2. l(Fig.A.1.1-2) and Fig 3-2.2(Fig.A.l.l-3) show an example of the front heater composed of two movable parts

Fig 3 -3 (Fig.B.17-3) shows a "patch" like front heater at the top.

Fig 3-4(Fig.B.17-4) shows a ring-like heater. Fig 4(Fig.B.18) shows an example of heating by air convection. Fig 5(Fig.B.19) shows an example of heating by radiation Fig 6-l(Fig.B.20-l) lists some heating methods. Fig 6-2(Fig.B.20-2) lists some heat sources.

Fig 7-l(Fig.B.19-l) shows an example of how coolers can be used to lower air temperature inside the drip chamber. Fig 7-2(Fig.B.19-2) shows how to dissipate heat from the drip chamber.

Fig 7-3(Fig.B.19-3) shows how traditional liquid heater can be modified to also heat specific area(s) of the drip chamber. We draw different parts of the heater as of different thickness to emphasize the concept of non-uniform heating.

Fig 8(Fig.B.22) lists some ways of generating low temperature.

Fig 14(Fig.B 22-1) shows the use of light sensor for improving signal quality for drip detection.

In its "Introduction"

The removal of splash droplets in general based on the same principle for removing dew droplets: keeping temperature of specific area(s) of the inner surface of the drip chamber above dew point could prevent vapor condensation and facilitate evaporation of existing droplets (either formed via condensation or by splash).

"Using the dew droplet removal/prevention methods in our present invention would significantly improve the signal quality particularly for the brightness variation class methods, as well as for other more sophisticated methods."

The essence of the dew droplet removal/prevention in US13897578 was to keep the temperature of specific area(s) of the inner surface of the drip chamber above dew point, and because the presence of dew droplets could interfere to some extent the accuracy of speed counting for all types of drip detection (in the electromagnetic spectrum including infrared ray, ultraviolet, laser, photodetector, etc; ultrasonic means), and that they all admit the same solution (temperature control), clearly US13897578 apparatus methods should be applicable in and out of the domain of using video/image processing methods.

Handle dew droplets by image processing methods

In image processing, just as in any other signal processing application, we always want to have signals of the highest quality and would like to remove noise as much as possible. Images of the IV chamber sometimes contain small dew droplets staying on the surface of the drip chamber, and when trying to identify the actual forming/falling drip we need to distinguish the forming/falling drip from these dew droplets.

This problem is illustrated in Fig.B 1-1 and Fig.B 1-2. Fig.B 1-1 shows dew droplets on the surface of the chamber that is closer to the camera, but the largest drip (can be identified using connected component methods in US12804163) has not yet come to the area containing dew droplets; however in Fig.B 1-2, when the falling drip comes into the "dew region", because the dew droplets are on the chamber closer surface closer to the camera (called "near/front surface" from here, and call the other surface which is farther to the camera the "far/back surface" from here), they could partially block image of the falling drip. As we see in Fig.B 1-2, even if we could successfully identify the drip location, we might either calculate a larger drip because the connectivity criteria in US 12804163 would merge it with the surrounding droplets, or get a smaller one because parts of the falling drip have been bitten/cut by the dew droplets. It is also possible that the remaining visible area of the falling drips becomes so small because of the blocking or "bitten" effect of the dew droplets so we might mistakenly identify another dew droplet as the largest connected component and hence the drip location.

Although in US12804163 publication [US 2012/0013735 Al] paragraph [0104]-[0108] we have already discussed the essential of the problem: Do a few problematic points invalidate the frequency estimation (Fourier analysis in US12804163, numerous others in US 13356632) algorithm? And the answer was that the few noisy points would not change the general periodicity of the signal so that frequency estimation algorithms could always recognize the correct period count. The conclusion has also been experimentally verified by the numerous experiments in US12804163 and US13356632, among which many include the "problematic" signal point of US12804163 Fig 3D.

We also show a real image of the dew droplets in Fig 21-4.2. This drawing was a photograph of our device's user interface (see Fig.B.21-1 to 3 and related discussion), and we display what the camera sees on the LCD screen so that patient/nurse could also monitor the dripping speed themselves and compare with the device's result. On the right side of the camera window we see about five dew droplets, but none of them is comparable in size with the forming/falling drip. In fact, in our experiments we have never recorded a case when the dew droplets interfered with the forming/falling drip identification using our connected component algorithm (see US 12804163).

In this application we present some additional processing methods that could further improve our result.

Comparing with the forming/falling drips, the dew droplets change their size and location rather slowly. The content in the image sequence (video) due to the forming/falling drips are the fast-changing elements, and the dew droplets are the slow -varying background. A host of techniques can be applied to separate fast-changing information from the slow- varying background. For example we could:

1. Compute only the difference.

We do this by first capture a frame and use this frame as the "base". If these images contain dew droplets, subsequent images taken shortly after it will also contain almost the same droplets as in the "base", and even if there are changes like disappearing or merging of some of the droplets, these changes will not be so significant as long as we keep the time frame (within how long a time frame after the "base" do we take image and compare with the "base") short. Therefore in general we could assume the background as static, and subtracting from each subsequent image the "base" yields only the difference signal from the "base", which in general would also be a periodic signal.

2. Use averaging to get the "base". One drawback for randomly taking an image as the "base" is that the "base" might happen to be an image which contains a forming/falling drip, as in the case of Fig.B.21-4.3. There are different ways of computing the "difference". Because "base" Fig.B.21-4.3 contains a large forming drip, if after taking the signed arithmetic difference between a later / and I_base, which is / - I_base,

(1) If further take absolute value | / - I_base |, we might end up always having the large forming drip area in Fig.B.21-4.3 the largest bright area, which could lead to the wrong identification of a nearly constant drip location as in "base" Fig.B.21-4.3.

(2) If we truncate the negative part for each pixel pair's difference, then we could still get a periodic signal which is amenable to frequency estimation.

Nevertheless taking images containing large bright drips as in Fig.B .21-4.3 as the background does not always seem like a logically impeccable method. To ameliorate this, we take a sequence, say 15, of consecutive frames, sum and then average. Since drip change its location across frames, then even for the maximum grayscale value 255, after /15 it becomes 16; and even if during the forming of the drip the position remains almost constant for a number of frames, say 5, dividing by 15 would still bring the area's (near the dripping mouth) grayscale level down to 1/3. In all cases after the averaging we would have the static dew droplet areas remain almost unchanged, but moving contents significantly darkened. By averaging we always get a better background than randomly taking an image.

If we want to get perfect signal quality, another approach is to prevent the dew droplet from forming so in image processing stages or remove them so that we get clean images from the beginning. The dew droplets form on drip chamber surface only when the surface temperature is EQUAL or LOWER than the liquid vapor's dew point. Dew point is associated with relative humidity, and as the relative humidity increases, dew point rises and get closer to the current temperature. Therefore if we could keep the temperature of the inner surface of the drip chamber above the dew point, no dew droplets would be able to form on the surface.

Fig.B .19 illustrates one method to achieve this. Preferably all elements in the drawing should be partially or fully enclosed in a housing to provide a "cleaner" low-noise environment for the drip sensor (whatever type), which is particularly important for sensors (transmitter-receiver pair) using electromagnetic spectrum signals because energy from external environments could interfere their detection, and we recommend enclosing the elements with a housing. However we did not include housing in Fig.B.19 here to emphasize that the temperature control apparatus and methods are independent from the housing aspect.

The transmitter can be seen clearer in the perspective view on the left and on the upper of the top view we see the flatter part being the receiver. These are merely conceptual drawings and do not represent real geometry (please also refer to "Drawings - Figures" illustration). In all views we see the drip chamber is being wrapped by two bended sheets on the front and back surfaces. Those wrappers are actually heaters providing local, rather than global, heating to the drip chamber. There is apparently windows intentionally cut in the middle of both the front and back wrapper (heater) with an obvious purpose of not to block "line of sight" of the transmitter-receiver (or sensor). The heat would be applied to the outer side of the drip chamber from the inner (concave) side of the wrapper, reaching the inner side of the drip chamber surface and also by convection (albeit slow on plastic) to the exposed/windowed area. As long as this applied heat keeps the windowed area's temperature above dew point, no dew droplet will be formed and we will always have a clean view. Depending on the type of transmitter-receiver (or sensor), we might optionally cut window in one of the surfaces.

The specification of US12804163 described in detail why a windowed area would suffice for drip speed measurement. Please refer to that for more information.

Similarly, Fig.B.17-5.1 and Fig.B.17-5.2 show different views of a back surface heater, corresponding to the annotated part of Fig.B .16-2. Using a back surface heater to keep some part of the back side of the inner surface of the drip chamber above dew point could also prevent dew droplets' formation on that area.

Combing the front and back surface heater, we could completely remove the dew droplets shown in Fig.A.1-1, Fig. A.1-2 and Fig.B.21-4.3. Combining the front and back surface heater with the illumination techniques discussed in US13356632, we would signals of almost perfect quality and a guaranteed rock-solid reliability for a medical application.

It should also be noted that we did not specify that both the front and back heather would be simultaneously required. As having been shown by the real image in Fig.B .21 -4.3 , in many situations (depending on liquid type, drip chamber material, temperature, environmental light, camera lens type, etc.) even if dew droplets exist they are still negligible and would not affect the detector, so the implementation could use just the front or back heater to remove some of the possible dew droplets and leave the remaining few to the treatment of signal processing algorithms.

That we are opening a window is based on the presumption that in generally metallic (nontransparent) material will be used for heating due to their good heat conductivity, however if transparent materials can be found which also has acceptable heat conductivity, it can also be used and the window would not be needed.

The shape of the window and the outline of both the front and back heater are also illustrational. Any reasonable shape can be used in real implementation. Please refer to section "Experiment and Calculations are important" for more information.

Nor is there any requirement that the front and back heater must be separated. We separate them only to make the concepts clearer, and in real implementation one could of course choose whatever combination or make them into an integral whole, as long as the same effect (keep specific area's temperature above dew point) can be achieved.

One might worry whether it would be possible for dew droplets to form on the top inner surface of the drip chamber and flow down to the windowed area (and the corresponding area on the back inner surface). We could add heating directly to the top surface to make it hotter than the dew point, as illustrated by the patch- like heather in Fig.B.17-3. The size of the heater in Fig.B. 17-3 is also purely illustrational and the actual dimension needs to be determined by experiment and calculation. Of course, the top patch in Fig.B .21-4.3 can also be put on the back side. The necessity of top "patch" like in Fig.B. 17-3 for preventing dew droplets from forming on the top could only be known after knowing the exact heat/temperature distribution of the drip chamber, please refer to section "Experiment and Calculations are important" for more detail.

Fig.B. 17-4 shows a "ring" heater surrounding the tube. As long as it can dissipate enough heat to the area of the inner surface to make them hotter than the dew point, it can also be adopted. We include it simply as an example to show the variety of shapes and arrangements the heater could be built like. For the "ring" heather, as long as the transmitter- receiver (or sensor)'s "view" window (see US12804163) does not stride or overlap the ring area, it would not cause any problem.

In building a real product one has to consider problems like how the drip chamber could easily be inserted/put into the device. A front heater like in Fig.B.16-1, Fig.B.16-2, Fig.B.19-1 might need to be moved away first before the drip chamber can be put in, in order to make the use easier we could divide the front heater into two halves, and use simple mechanical structure (for example, hinges driven/rotated by gears) to cause it to open/close before and after putting in the drip chamber, as shown in Fig A.1.1 -2 and Fig A.1.1 -3.

It is obvious that these mechanical alterations, just as shape of the heaters, are unimportant comparing to their function in keeping temperature of the specific areas of the inner surface above dew point. There are numerous ways to achieve the same effect as in Fig A.1.1-2 or Fig A.1.1-3 but the essentials would be the same.

Experiment and Calculations are important

FromFig.B.16-1 to Fig.B.17-5 we give no specification on the shape, size, multiplicity (how many) or other parameters of the heaters. In real implementation we face some constraints:

1. Excess heat causes humid air temperature to rise and might affect dew point. Although for back heaters like Fig.B.17-5.1 it is guaranteed that drip chamber inner surface will be hotter than air because it is in direct contact with the heater, for windowed area like in Fig.B.17-1 and Fig.B.17-2 the conclusion is less certain because the windowed area is heated by weak conduction of drip chamber's plastic material.

2. From power consumption perspective we should also minimize unnecessary power used on heating. Because we use the tube-constricting mechanical systems as disclosed in US 13019698 and US13356632 rather than peristaltic pump, the power consumption of the whole IV monitoring and controlling device could be made very low, and in this situation the energy dissipated on heating could be significant when comparing with other parts.

In designing the real product we need to strike a balance between the need of keeping inner surface ' s specific areas ' temperature above dew point, and the considerations above. To reach an optimal design one might need to resort to

(1) Theoretical calculation

(2) Computer simulation

(3) Experiment, such as analyzing heat distribution by thermal imaging Only after getting quantitative results from the work above could we know the optimal shape, heating temperature, as well as other parameters. Whether we would need the "patch" as in Fig.B. 17-3 to prevent dew droplets' formation on the top is also a question that can only be answered after knowing the exact heat/temperature distribution.

Convection, Radiation and Advection

The heaters disclosed above all have direct contact with the drip chamber and therefore heats by conduction. The drip chamber surfaces can also be heated by

1. Convection

a. Air: as in Fig.B . 18. The heat source can be of any type and the heat source drawing is only an iconic symbol. The fan is optional and is for facilitating air convection.

b. Liquid: such as using liquid to carry heat from a source to drip chamber surface.

2. Radiation: as shown in Fig.B. 19(1) and Fig.B. 19(2). The heat source drawing is also an iconic symbol and can represent any heat source capable of radiating heat, for example a miniature infrared heater.

Fig.B. 19(1) is fromUS 13897578, heating by radiation; Fig.B. 19(2) as is from US 13903924, heating by radiation, the radiation source can be on either side of the transmitter/receiver pair/imaging device (sensor), and can for example be a miniature infrared heater.

3. Advection: It is also possible to implement advection (by air or fluid) to transfer heat to the drip chamber surface with some components.

4. Heat pump: one can also use various types of heat pumps to transfer heat to the specific areas.

For these three methods, heating the back surface is not as easy as by direct contact conduction. The

calculation/simulation/distribution of heat distribution could also become considerably more difficult than the direct contact conduction heater method, and more effort will be needed in getting the optimal result.

Monitoring and controlling temperature

There are different ways for setting the desired heating temperature. For monitoring temperature of the drip chamber surface, or possibly even the inside, one could use thermocouple (using Seebeck effect, etc.), thermal imaging, temperature sensing resistor, thermistor, thermometer or else; for controlling temperature one could use a thermostat (electrical, analog electronic, digital, mechanical, etc.) or else. It should be noted that the choice among these methods, or even future techniques, is unimportant, the important thing is to keep temperatures of specific areas of the drip chamber's inner surface above dew point. Fig.B. 20-1 summarized the heating methods we have discussed so far. Heat source

A vast variety of heat source can be used, the specific choice being unimportant. One should always note that what is important is the purpose of keeping temperatures of specific areas of the drip chamber's inner surface above dew point.

Fig.B. 20-2 lists some common methods of heating:

1. Ordinarily one can use Joule heating.

2. Oil or other material can also be burned to generate heat

3. The heat of the battery, or heat generated on the PCB board / by components can also be directed the heat the drip chamber.

4. Thermoelectric effect, including using Peltier effect / Peltier module.

5. Other heat sources (not shown in Fig.B. 20-2).

If (3) above is used one has to ensure that the PCB board / components / battery be hot enough and properly preserve the heat when directing it to the heating location. And whatever heat source is used, one must do the

calculation/simulation/experiment properly to obtain the optimal parameters.

Cooling, and Dissipating Heat and combine Heating/Cooling

All the heating methods described so far have cooling as their duals (opposite/complement). This is because our ultimate goal is to keep temperatures of specific areas of the drip chamber's inner surface above dew point, to achieve these we can either increase the temperature of these inner surface areas, or lower the dew point. The dew point can be lowered by lowering the humid air temperature.

Therefore if instead of explicitly transferring heat to those drip chamber surface areas, we may remove heat from the air, or from the liquid which will in turn lead to the lowering of the temperature of the air, and we somehow maintain (or raise, or lower it but keep it still higher than air/liquid) temperature of specific areas of the inner surface of the drip chamber, then they will still be above the dew point.

The wrapper arrangement in Fig.B. 19-1 looks exactly like a dual (opposite/complement) of Fig.A.1.1-1. Whereas in Fig. A.1.1-1 heat is explicitly transferred to windowed area and back surface area, in Fig.B. 19-1 heat is removed from the two side surfaces which could lower the air temperature inside the chamber. For the drip chamber surface area facing the camera, as well as the it back (far) side, because the heat conduction of the drip chamber's surface is slow, it is possible that the temperature of these areas decrease slower than the air inside the chamber, and in this way we have successfully keep these areas' temperature above the dew point. If we would like to cool the liquid, we could move the patch-like structure in Fig.B . 17-3 to the bottom and it would effectively lower the liquid temperature down, and consequently temperature of the air above the liquid.

Because the dual (opposite/complementary) relationship between cooling and heating, all the heating methods listed in Fig.B. 20-1 and/or discussed above, as well as all arrangements from Fig.A.1.1-1 to Fig.B. 19, can be used on cooling after straightforward modifications.

Instead of cooling by which we mean external energy (work) is required to transfer thermal energy away from the drip chamber, we might also use dissipating process which happens spontaneously under the second law of thermodynamics In Fig.B. 19-2 the drip chamber is clamped by two pads which can actually be the drip chamber's fixture, and they are preferably made of material of good thermal conductivity such as metal. These two pads effectively absorb any excess heat of the drip chamber if its temperature is higher than the two pads. This is especially useful when the drip chamber is being heated by some method so that it has higher temperature than surrounding, and even without external energy (work) the much of heat, at least on the two lateral sides as shown in the drawing, will be dissipated through the two pads (absorbs), and the purpose of keeping the temperature of the specific area(s) of the front or back inner surfaces to be above dew point can be achieved.

We emphasize again that the shape of heat dissipating mechanism is unimportant and we use the flat fixture-like structure in Fig.B. 19-2 merely as another example to show the variety of shapes these elements could assume. Nor is there any requirement on the multiplicity of these elements.

It is also obvious that cooling/dissipating/heating can be applied simultaneously to create the relative difference so that the temperature of specific areas of the inner surface of the drip chamber is above the dew point.

Generating cold temperature

Fig.B. 22 lists common methods for generating low temperature. These method include

1. Vapor-compression refrigeration (Refrigerants)

2. Absorption refrigeration

3. Air cooling

4. Reverse Stirling cycle heat engine

5. Thermoelectric effect (Peltier effect, Peltier module)

All methods can be used, however method 5's implementation relatively is the easiest among the listed method above. No matter whatever cooling method is used, one must do the calculation/simulation/experiment properly to obtain the optimal parameters. Applicability

Our fundamental method of keeping the temperature of specific areas of the inner surface of the drip chamber above dew point could improve signal quality for all types of periodic measurements including those disclosed in

US12804163 and US13356632, no matter it is trajectory (height) based, drip size based, brightness variation based or others. For the methods disclosed by Cheng US 12/791,885 Intravenous Drip Monitoring Method And Related Intravenous Drip Monitoring System, whose basic idea is equivalent to our average gray level measurements in US12804163 (Fig 41 and Fig 4J and the corresponding specification text), the removal of dew droplets is actually more important because for this class of brightness variation methods we usually do not get the information as rich as in the trajectory (drip height) or size measurement, and these brightness variation measurements are more susceptible to interferences from the dew droplets.

For methods not using video/image processing for drip rate measurement, for example detecting drips based on infrared ray, our dew removal/prevention is even more important. For example Fig.B. 22-2 shows the ideal infrared ray receiver signal in which the valleys correspond to moments where falling drip reflected/blocked the ray, and the period in the signal could be easily determined by even the simplest method; Fig.B. 22-3 corresponds to the situation where there are dew/splash droplets (not in the trajectory of falling drips) during the middle of the sampling frame, and clearly these dew/splash droplets by themselves blocked the ray(s) which resulted in the basin shape of the middle. With infrared ray transmitter-sensor (or sensor; or UV, ultrasonic or other means) there is no guaranteed method to handle these cases with guaranteed reliability.

With the addition of dew removal/prevention apparatus and methods above:

1. Dew droplets will not form

2. Splash droplets will evaporate and cannot stay for an extended period of time. If we detect abnormal pattern in the signal like in Fig.B. 22-3, we might wait for some time or raise the temperature (or plus cooling/dissipation to increase the relative temperature difference) of the heating system to facilitate the evaporation of these droplets.

Differentiate from Liquid Heater

There are some prior arts for heating the liquid. For example in application WO2009109559 "Intravenous Fluid Monitoring Device", which uses "[0019] resistive heating material" which "[0020] in one embodiment composed of thermally conductive metal-alloy tubing". However the purpose of this device is to control the fluid temperature of the drug liquid to what is appropriate to the patient such as what WO2009109559 described" [0019] from 20° to 38° ±0.5°".

Whereas the disclosure in WO2009109559 uses a housing to enclose both the dripping tube and the heating structure so the heating is applied at the drip chamber location, in the market there are also attachable devices which encloses/wraps only segment of the thin-tube, typically at the end near the injection site of the body. These designs have the advantage of presumably higher energy efficiency which would result in lower power consumption. A common characteristic of these devices is that their primary heating energy is directed to the liquid, whose difference with our disclosure on heating (cooling, dissipating heat) particular areas of the drip chamber surface should be instantly recognized from Fig. A.1.1-1 to Fig.B.19-1 and the specification. The reason that our design does not heat liquid is:

(1) Water constitutes the majority of all types of drug solution and it has a very high specific heat capacity of 4.18J-g^_1K^_1. If we were to heat some liquid from 20° to 25° (not even body temperature 37° !) for lOOOmL solution, then the energy required is

4.18-1000 (water design lg/mL)^■ 5 = 20900 J and if we are using lOOOmAh battery, its energy storage is

3.7VTAh = 3.7Wh = 13320J, so even the full battery capacity of a 1000mA battery might not sustain the heater usage alone to the end of the dripping.

(2) More important, it violates our fundamental principle of keeping the temperature of specific areas of the inner surface of the dripping tube above dew point. If liquid get heated, it will soon heat the small amount of humid air above it inside the small enclosed space of the drip chamber, plus evaporating vapor at the almost same temperature, so consequently the dew point temperature also rises and could become very close to drip chamber surface temperature, in which case droplets could form, either directly at the "view"/" observing window" of the sensor/transmitter-receiver, or elsewhere which could flow down along the surface to the "view'V'observing window" and cause interference to the detection signal.

For the type of heater which is fixed in a housing which encloses the drip chamber (either fully or partially, or even clamping/embracing the drip chamber only at some part where it could get in contact with the liquid), if we do want to use this type of heater to heat the drip chamber at the same time to keep the temperature of certain area(s) of the inner surface of the drip chamber above dew point, we obviously could NOT apply the heat uniformly, i.e., apply the same or heating temperature to the liquid as well as to the specific area(s) of the drip chamber. In order to keep the specific area(s) of the drip chamber surface warmer than the liquid, the heating element needs to give NON-UNIFORM heating in which the area(s) of the heating element(s) that is in direct contact with the specific area(s) of the drip chamber surface be HOTTER than the area of the heating element(s) that is used to heat the liquid. We emphasize again that this is the fundamental principle of modifying a traditional "liquid heater" to encompass dew

prevention/removal ability.

In one embodiment, this can be achieved by using a resistive heating element of non-uniform resistivity (or some specific geometric/dimensional configuration) so that the Joule heating power (I² R) is higher at specific area(s) than other; another embodiment could be using discrete heating elements to heat separately liquid and chamber surface's specific area.

Modification on liquid heating elements to encompass our dew removal/prevention functionality could be spotted/identified by: (1) See if the heating elements heats specific areas of the drip chamber surface that is not required for heating liquid alone.

(2) Examine to see whether heating elements of different type, property, parameters (resistivity, etc.) are used to heat liquid and the specific area(s) of the drip chamber surface.

(3) Use thermal measurement instruments, such as thermalcouple, thermal probe or thermal imaging, to find out the temperature distribution of the entire heating system and see if the specific area(s) of the drip chamber is being heated to higher temperature than the liquid area.

An example drawing in shown in Fig.B. 19-3 in which we draw heating elements of different thickness to signify nonuniform heating. The circular geometry is purely illustrational and in practice any reasonable shape/construction could be used, but all modifications to make traditional liquid heater capable of dew removal/prevention should use the principle that the specific area(s) of the drip chamber be heated to higher temperature than the liquid.

The heating system as disclosed in our present and earlier US13897578 application has no restriction on the heating temperature as long as specific area(s) can be kept above dew point. Hence, it is permissible to heat the specific area(s) of the drip chamber surface above normal body temperature 37°. Therefore in one embodiment we might specifically set the heating temperature of the specific area(s) of the drip chamber surface to a value equal above 37°, say 38°, 40°, 42° or higher, whereas for ordinarily liquid heating the need to heat liquid above 37° or 38° is extremely rare except for patients in severe conditions or for very special diseases. For many types of diseases/ailments, for example fever, heating drug liquid above temperature could only deteriorate their condition. Therefore, we added a specific limitation that our dew removal/prevention apparatus and methods allows heating specific area(s) of the drip chamber surface to above any reasonable (as long as not generating excessive heat for drip chamber and inside humid air) temperature above normal body temperature (37°~38° with tolerance) . Except for severe or particular disease patients, ordinary liquid heating should not attempt to heat liquid to above (or significantly above) normal body temperature.

The use of camera

Fig.B. 21-5 shows the schematic drawing that a camera is put inside the housing to capture the dripping process. One side of the housing is removed to reveal the camera and drip sensor/transmitter-receiver.

Fig.B. 21-6 shows the full housing, and on the external there is a display showing the video captured by the camera inside.

We emphasize that although a housing enclosing the detector inside it is very helpful in achieving good signal quality by blocking external light (and noise for acoustic/ultrasonic detectors, and others types of noises for their respective types of detectors), it is not however not imperative. As long as sufficiently -well signal quality can be achieved, a partially enclosing housing, or even without it, would be fine. The imaging device (camera) can be used in combination with non-imaging detectors with and without the presence of housing. The combination of a camera and drip detector (sensor/transmitter-receiver) solves many of these detector's inherent limitations:

1. Non-imaging drip detecting methods (IR (infrared ray), UV, RF, laser, other electromagnetic spectrum, ultrasonic, etc.) are typically susceptible to external environment's interference. For example, many IR drip counters are positioned toward the drip chamber in the open air and external light could cause problems to their detection. The reason they are not enclosing the drip chamber and drip detector inside a housing is because if they do this, patient/nurse won't be able to see the dripping process and tend to be skeptical of the dripping process. Use a camera to route the video to external display solved problem while could also allow a much better detecting environment. However, camera can also be used without a fully-enclosing housing, or even without such housing.

2. The camera affords much richer enhancements include:

a. Dew/splash detection

b. Drip rate measurement

c. Accurate drip size measurement

d. Capture dripping process inside the enclosing housing to allow display by an external screen.

We elaborate on the above. Dew/splash detection

In an enclosed environment as shown in Fig.B .16-1 the relative position of the camera and the drip chamber is fixed. Position of the dripping mouth in the image is always known, and relatively static bright spots near the dripping mouth can be easily identified and handled by:

1. The "difference" and "averaging" methods in our US13897578, section "Handle dew droplets by image processing methods".

2. Connected component based methods, for example, as in our US 12804163.

3. Edge detection based methods. For example, the Canny edge detector or Laplacian.

With the knowledge of the presence of dew/splash droplets, the problematic signal illustrated in Fig.B.22-2 and Fig.B.22-3 could be corrected and then used to calculate the drip rate accurately.

In addition, after detecting the presence of dew/splash droplets, we might also increase the temperature (can combine with cooling and heat dissipation methods) to facilitate their evaporation. This should be able to remove dew/splash droplets for the specific area(s) quickly and restore the normal clear detection window. Drip speed measurement

Drip rate measurement has been discussed at length in our US12804163 and US 13356632 applications, and could be with measurements from non-imaging methods for cross-verification.

Accurate drip size measurement

An inherent limitation of non-imaging drip detectors is that they cannot measure the drip size accurately. Drip size is affected by many factors:

among other factors.

Therefore knowing the drip alone is not sufficient to calculate the volumetric rate, even with some knowledge of the tube material and/or specifications from the IV set manufacturer which gives the relationship such as how many drips equals to one millimeter. In order to achieve high volumetric measurement accuracy, imaging device is necessary.

Prior arts for measuring drip size from imaging devices include:

1. Kai Tao, US12804163 IV Monitoring by Video and Image Processing

2. Kai Tao, US 13356632 Image Processing, Frequency Estimation, Mechanical Control and Illumination for an Automatic IV Monitoring and Controlling system

3. Zandifar, US7783107, Characterization of aPrinted Droplet.

In addition, because the drips in their forming phase (near the dripping mouth before breaking) are hanging down under both the effect of gravity and surface tension, the known relationship of Young-Laplace equation can be incorporated into calculation to yield more accurate result.

Particularly when measuring drip size attention has to be paid on the imaging device (camera) 's exposure time for each frame. We take digital cameras for example. Digital cameras typically define exposure time as a multiple of line scanning time, so if we set N in the exposure time control register, the exposure process will take an interval equal to the scanning time of N lines. The larger N is, the brighter the image, however for fast moving objects this also mean that the image captured might not correspond to a single position, but is actually the momentary trajectory of the object (droplet) during the exposure time, and would appear longer than the actual object, which causes serious problems to accurate size measurement. Therefore two measures are necessary:

1. We need to set a low exposure time, like by setting the N above low for a digital camera.

2. Correspondingly we need to increase illumination to compensate for the reduced exposure time in order to obtain drip images of the needed brightness level.

By combining camera and image/video processing technique with non-imaging drip detectors, we were able to achieve high accuracy in both drip rate and volumetric rate measurement. To our knowledge this combination is not disclosed or existed in prior arts.

Display dripping video on the display

In our series of applications (US12825368, US12804163, US13019698, US13356632) we enclose the drip chamber inside the device and blocks external lights to create an ideal shooting environment for the computer vision system. In fact an ideal environment is also important for non-imaging detectors, for example an infrared ray detector. However from the patient/nurse 's standpoint, without being able to see the actual droplets coming down from the dripping mouth they tend to be skeptical on the calculated speed as being displayed on the external display (see Fig 1 in US12804163). Other applications such as Cheng 12/791,885 did not mention the use of an external display.

CN201110955Y of Enmin Song displays only numeric data, but not the video of the dripping process as in our present disclosure.

We believe a visual display of the dripping process inside the device is essential to our user's experience and to their confidence with this device. Therefore as in the design drawing of Fig.B. 21-1, we specifically dedicate an area on the display for showing the video of the dripping process. The video is captured by camera as shown in Fig.B. 21-5 and the screen is shown in Fig.B. 21-6. Although preferably the display is a touchscreen LCD through which all input/output can be exchanged graphically and interactively, there is no inherent requirement that LCD be the only choice (CRT display may also work; new display technologies might soon emerge, etc.). What is important is that we display at least an area of the camera's view on the display which contains at least some part of the drip's forming/falling process, allowing the user to see the actual dripping process and verify the machine's counting themselves.

For aesthetic purposes we might also represent some part of the dripping chamber by UI graphic just as shown in Fig.B. 21-3, and display only a small window of the camera's video which contains the enough information to the human observer (user) (Fig.B.16-0).

Fig.B. 21-4.1 to Fig.B. 21-4.4 show four frames of video being displayed on the LCD screen of our real device. The formation of the drip in Fig.B. 21-4.1 to Fig.B. 21-4.3 as well as the eventual fall in Fig.B. 21-4.4, are very clear to the user and they could easily count the speed themselves and compare with the algorithm result (shown as 80 drips/mean in the callout window on the right). Because inside the microprocessor the display and video input module typically use different buffer and memory space, when implementing this one needs to properly copy the input video frames to the display output buffer, and the details would depend on the specific choice of the processor and peripheral ICs.

It is also possible to use separate display devices, for example one LCD (CRT, etc.) module to show only the video of the dripping process, and other modules to show numeric monitoring information (possibly by simpler and cheaper display technology). This type of arrangement is shown in Fig.B. 21-2.

What is important here is provide a means to the user to monitor the actual dripping process of the drip chamber enclosed inside the device so that they could verify the displayed counting themselves and gain confidence to the device.

Another implementation is to open a window on the devices housing, which can either be a permanent opening or a window that can be covered by a lid/cover (could be a fully or partially transparent window), and allow the user to monitor the inside by seeing into that window area (possibly after moving the lid/cover). This arrangement is also shown in Fig.B. 21-2.

Yet another implementation is to use a one-way mirror on the device housing. The purpose is to allow the dripping process to be seen from the outside whereas minimizing the interference from outside. Because the use of one-way mirror requires a strong contrast between the illumination of the two sides, we might need to use strong transmitters (hence strong signal/noise ratio) and strong illumination inside the housing to achieve the effect.

Light sensing

The accuracy of drip detector / various monitoring apparatus can be further improved by incorporating a light sensing component. Light sensing components can use different principles and the exact technology is unimportant comparing to their functionality. Common components for such purpose might be called "ambient light sensor" or "photo detector", and are from various manufacturers like Intersil, Vishay and AMS.

The chief reason that drip detectors (both imaging and non-imaging) could be susceptible to external interference is because energy from them can become strong noise in the signal. In dark room the infrared ray detectors could be accurate because there is little interference, but under sunlight the environmental light intensity might be comparable to the transmitter signal strength, so the signal/noise ratio is low, and detection would be difficult; abrupt environment change such as opening a door to introduce outside light into a dark room might also trigger false detection.

A light sensing component detects these changes. For

1. Abrupt change, it can simply drop the current detection or interpolate from previous drip rate, and in general they won't cause significant problem because abrupt changes are rare.

2. Strong interference:

a. If acceptable (e.g. signal/noise (S/N) ratio high enough), then do one or both of

i. Increase transmitter signal level, which increases S/N ratio. ii. Change signal processing algorithm, such as changing threshold level

b. If not acceptable, then alarm.

Light sensing can also be done by the camera, since camera itself is essentially composed of arrays of photo detectors. The measurement of light level can be done in software, or many of today's camera modules already incorporated related detection/compensation algorithms in the circuit, and the CPU could reach corresponding registers of the camera module to know information related to the current illumination environment.

The operation of a light senor in conjunction with drip detection is shown in Fig.B. 22-1.

And in "Abstract" of US13897578 and US13903924:

When analyzing video frames captured for monitoring the IV dripping process, dew droplets could exist in the image. We discussed image processing methods to remove the dew droplets from the background, including computing the difference between frames and averaging to get a proper background image. We also discussed various methods to keep the temperature of some areas of the inner surface of the drip chamber to be above the dew points in order to prevent dew droplets' formation or to remove them. In the end we showed how video of the dripping process could be shown on external display(s) for devices enclosing the drip chamber inside.

Dew and splash droplets cause problems to IV monitoring devices of all kinds and we discussed apparatus and methods of temperature control to remove them or prevent dew droplets' formation. We also described how to combine imaging and non-imaging devices to get more accurate drip speed and size measurement, and the use of camera to capture the dripping process and display it on an external screen. In the end, we described the use of light sensing components for improving signal quality for drip detection.

[END: US13897578 and US13903924 contents]

Already described in Applications US13897578 and US 13903924: Fig.B.22 lists common methods for generating low temperature. These method include

6. Vapor-compression refrigeration (Refrigerants)

7. Absorption refrigeration

8. Air cooling

9. Reverse Stirling cycle heat engine

10. Thermoelectric effect (Peltier effect, Peltier module, etc.) All methods can be used, however method 5's implementation relatively is the easiest among the listed method above. No matter whatever cooling method is used, one could do the calculation/simulation/experiment properly to obtain the optimal parameters.

Thermoelectric cooler have efficiency about 8%, and higher efficiencies between 20% and 30% have been reported. With temperature feedback mechanism thermoelectric cooler can achieve 0.0 IK accuracy (for consumer products, higher accuracy for specially tuned applications), and even separated by a layer of drip chamber material, we can still expect a satisfactorily accurate temperature control be achieved.

In addition, the cooling might utilize (include, "open-ended")

1. All methods in refrigeration

2. Water cooling

3. Heat pipe

4. Heat sink

5. Synthetic diamond cooling/heat sink

6. Phase change material

7. Electrostatic fluid acceleration

8. Heat spreader

9. Vortex tube

10. Stirling engine/machine based cooling

11. Fans

12. Piezoelectric pump

13. Liquid submersion

14. Electrostatic air movement and corona discharge effect cooling

15. Phase-change cooling

16. Liquid nitrogen

17. Liquid helium

18. Air connective cooling

19. Liquid convective cooling

The cooling methods might be used simultaneous with any heating mechanism disclosed in application from 056090 to this application as the single objective is to keep the viewing window (front and/or back) hotter than dew point.

The temperature effect of cooling, defined as measurable when heating mechanisms are not operational, is (to elect if conflicting with other applications) from 0.00 IK, down in 0.0 IK decrement, until 40K or 273.15K (freezing point), whichever is higher, and is defined as the change for either the max, average, or min temperature, for liquid in drip chamber, or liquid in the tube before entering the drip chamber. For liquid in the drip chamber, the volume over which the temperature measurement and max, avg, or min calculation is done is define to be the whole, or lowest 1/3 current volume, or middle 1/3 of the current volume, or the top 1/3 of the current volume.

For liquid being heated in the tube above the drip chamber, the volume is defined:

1) If there is only one integral heating length (one segment, not separated segments), the volume over which the calculation is done is defined to be the entire length being heated, or the lowest 1/3 part of that length, or the middle 1/3 part, or the top 1/3 part.

2) If the heating is done over disconnected segments, the {entire, lowest 1/3, middle 1/3, top 1/3 } is defined either for each individual segment, or any combination of segments up to the entire aggregate length.

To further increase cooling efficiency we may adopt thinner drip chamber material either at cooling site(s) cooling liquid within the drip chamber, or at tube cooling length above the drip chamber, and the limitation was given in

"^Thinner drip chamber/tube for increased cooling efficiency" .

In Fig.B.27-1, the outer of the liquid area (up to liquid surface height) is wrapped with cooler at constant boundary condition 292K, and we see in steady state the isolines of gas inside the chamber protrudes up. From this we see that at chamber surface - gas interface, chamber surface always actually is in contact with gas at a temperature of a lower isoline - lower temperature, hence condensation is prevented.

Fig.B.27 -2 repeats the experiment with boundary condition = 295K. The external environment temperature is 298K.

The concave cooling window in mid of Fig.B .28-1 is simulated in Fig.B .27-3 with two windows together (structure and dimensions in Fig.B .27-3(0)) pumping/pulling out 0.05W heat from outside of the PVC (not directly from gas), and the cooling was stronger than needed as shown by the 271.22K lowest temperature; with lower power we can easily achieve mild cooling. 0.05W heat removing power is very easy to achieve with modern Peltier cooler (of course allowing other types of cooler).

Fig.B.27 -4 changes boundary condition to 292K at the two sockets' windows in contact with PVC.

In both Fig.B.27-3 and Fig.B .27-4 we see at viewing window (sockets are lateral, viewing window contour/isolines at the orthogonal cut plane) chamber surface always hotter than gas, hence no condensation will form on viewing window.

More calculations has been done in Fig.B.27-3 for Q from -0.05W to -0.00001W (figures for -0.0001, -0.00007, - 0.00005, -0.00001 not shown), all showing that temperature of gas increases from cooling windows to viewing window, in which case no condensation will form. For the model in Fig.B.27-3(0), the volume of gas is 3.737mL. By dividing this cooling power rate with this gas volume we obtain a per mL cooling rate (steady state or average-steady state), all of which guarantee that the viewing window (front and back, due to structural symmetry) will not form condensation. watt

At ^'

-0. 05 - 13 , 3782

-0. 02 - 5 , 35127

-0. 01 -2. 67564

-0. 007 - 1 , 87295

-0 , 005 -1 , 3.3782

-0 , 002 -0 , 535127

-0 , 001 -0 . 267564

-0 , 0005 -0 . 133782

-0 , 0002 -0 . 0535127

-0 , 0001 -0 . 0267564

-0.. 00007 -0 . 0.187295

-0.. 00005 -0 . 0.133782

-0.. 00002 - 0 , 00535127

-0.. 00001 - 0 . 00267564

As discussed in 6. Extension, the temperature sensor used to measure IV tube, solution bag and the monitoring window (even if radiation is not used or it is not the chief agent of heating when heating/cooling is undertaken by other means including but not limited to contact cooling/heating) is also introduced here. The structure has been proposed from US13897578 and CFD simulation given from IB 2013/056090 application and the temperature claims these priorities.

Coisc&ve drip efcasssber issss:si.ed c«»fer

To increase the efficiency of contact thermal conduction (both heating and cooling) we propose concaved "sockets", either/both at liquid area and gas area of the drip chamber so that heating/cooling device can be "plugged" inside to have more direct contact with liquid/gas. There can be a multiplicity of these sockets as shown in Fig.B.28-1 schematic. In the middle of Fig.B.28-1 the lateral concaved/depressed windows can host for example Peltier cooler which effectively cools the gas to prevent condensation.

To further increase contact thermal conduction we propose that the "sockets" uses material thinner than bulk of the drip chamber. The thinner material applies also to where explicit depressed "sockets" are not used as in Fig.B.16 to Fig.B.19.

Numerically, we propose: 1. Absolutely, the thickness at contact thermal heating/cooling sites be more than 0.05mm thinner than thickness at the "torso" of the drip chamber, by which we mean excluding caps and tube connectors at the top and bottom. For drip chambers assuming shapes as in Fig.B.25 and Table.B 4-3 to Table.B 4-3 the meaning of "torso" should also be self- evident. For the most common 0.5mm-thick PVC drip chamber 0.05mm is 10% difference which is already very large. However, if such a limit falls in the tolerance of manufacturing, we propose increasing in 0.01mm step from 0.5mm. Any material at heating/cooling sites/sockets thinner for more than a value in [0.05mm, 0.5mm] range is deemed as being used specifically to facilitate heat conduction.

2. Relatively, we propose at heating/cooling sites the thickness be more than 10% thinner than at the torso, and allow electing from [10%, 30%] in 1% steps if overlapping with manufacturing tolerance.

3. Material- wise, we propose using a different material (different defined by chemical composition analysis, by FTIR spectrum analysis, different physical property) other than other parts of the torso. Such material and the other parts of the torso are assembled/welded/linked as a whole, and serves to increase thermal conduction. It is analogous to "heat absorbing" drip chamber defined in ^Increasing absorbance, and if any of the "heat absorbing" drip chamber also material facilitates thermal conduction we also include for conductive heating/cooling.

The above "socket'Vdepression facilitate cooler's contact (via chamber) with internal gas. To facilitate the energy exchange between cooler (where applies, also to heater) and drip chamber and reducing thermal resistance, we propose:

1) Using grease (thermal grease, glue, silicon, oil, etc.), preferably purpose heat-conducting material between cooler/heater and drip chamber.

2) Instead of using conventional cylindrical drip chamber, the part of the drip chamber in contact with the heater/cooler is partially or fully flat (the flat is defined by using the rl and r2 definition below for the "flatter" viewing window).

To mitigate radius i?'s change's effect as discussed in §2.4 Image Simulation, we propose: The top-viewing profile of the drip chamber at one or more cut planes having either/or:

1. Straight edge segment, particularly in camera's active viewing area (defined by where either is viewable by the camera, or a subset of the complete viewing window where the active frame sequence is being updated, rather than static or less-frequently-updated "boundary" or margin).

2. Elliptical, polygon, rounded rectangle shape or any profile shown in Table.B 4-3.

3. Viewing window flatter than cylindrical case. Technically this is defined by : a. Taken the top-view cut plane profile at height corresponding to optical axis, or take several such planes within ±25mm height of the optical axis. For the planar cut profile, we define two radii:

rl: the radius of the profile's inscribed circle. This definition suits best for axial symmetric shapes. r2: First project dripping mouth to the plane, then draw a circle using the projected point as center, increase its radius until it touches the profile, denote this radius as r2.

If the absolute value of the (means the surface can be convex ) radius of curvature at the front (camera side) center (according to optical axis) is more than 3% (allow electing "more than" 4%, 5%, up to more than 100%, including intermediate values like 20%, 30%, 50%, 70%, increment in 1% step) larger than the derived either rl and r2 above, the we defined this planar profile as being a "flattened profile".

If in the profile set(s) of step a) we have identified any flattened profile, the drip chamber is defined as a "flattened viewing window drip chamber". The advantage of which is that flatter windows are less susceptible to shape change of the drip chamber.

In the last, we propose designs of the cooler/heater:

1. The cooler/heater being made of flexible material (one shall refer to classification on mechanical engineering literature or physics books, like The CRC Handbook of Mechanical Engineering, etc., for what material is flexible and what is rigid). The effect of the design is that the cooler can mate with chamber more closely, possibly under pressing/pushing forces.

2. The cooler/heater structurally having {pivot, gear, link chain, etc. } or likewise elements to allow deforming. The effect of the design is that the cooler can mate with chamber more closely, possibly under pressing/pushing forces.

3. The cooler/heater includes containers of bulk (not micro-scale) liquid (or colloid, emulsification, gas, jelly, etc., or inclusive as non-solid) which could deform. The effect of which is that deformation allows better surface mating to facilitate heat conduction.

To avoid that the contact pressing force would alter geometric (optical) shape of the drip chamber, we further propose: the contact between drip chamber and cooler/heater includes fibrous/cilium/antenna-like, soft/flexible material, preferably with resiliency (elasticity) (£f It , £f¾ , fl_i£A¾^"#tt#). The effect would be reducing the rigid/solid' s pressing force on the drip chamber to minimizing chamber deformation/

Instrument: AT AGO DR-M4 Multi- Wavelength

Abbe Refractometers

Wavelength: 546nm (filter ± lOnm)

Temperature: Celsius 20 1.5771

30 1.5552

35 1.5512

40 1.5449

45 1.5408

50 1.5341

55 1.5278

60 1.5188

65 1.5160

With AT AGO DR-M4 instrument we measured refractive index change of PVC drip chamber between 20° and 65° and in fact even for a change of 10° between 20 and 20 degree the index changes for over 0.02 (0.001 is already considered large for lens glasses). Therefore, we propose extending the simulation methods in part A and part C to temperature (by heating and by environment) induced n change. The methods are directly applicable as a parameter change.

Fart Cont nuation on Optical Correction

A Note on Camera Model

At the beginning of all the text below, we establish the equivalence between the "physical" type of camera model and the "mathematical" type of the camera model:

1) The physical type camera model corresponds to real lens parameters, including radius, coefficients of

aspheric terms, thickness, index of refraction of index, Abbe number of the lens system. Their corresponding tolerances might include tilting, decenter, spherical and astigmatism irregularity, deviation on radius represented either in length or by fringe number and so on. With strictly the physical model, corresponding between image space and object space is at the first place established by ray tracing through optical surfaces (or using PSF array /matrices).

2) The mathematical type camera model corresponding to the one mostly used in computer graphics / computer vision. The lens system can be modeled behaviorally by a fixed number of parameters which might not all correspond to those directly tangible physical quantities. For example, the mathematical model might include focal length which has a obvious correspondence to lens itself, but it might also include many terms which do have physical significances but was the result of sophisticated mathematical derivation. These mathematical models are frequently represented in the form of matrices / vectors /formulas and their use could greatly facilitate the calculation. The tolerances or ambiguities of these mathematical can be kept in well-controlled manner to achieve a high degree of accuracy.

Therefore at any instances below when we speak of ray tracing, either from object space to image space or vice versa, or we say "mapping", or we refer to "lens model", tolerances, they applies both to the physical model and mathematical model. Any ray -tracing we therefore perform implies its equivalence using the corresponding mathematical model (such as by matrix/vector multiplication). When we estimate the internal lens parameter tolerances in ^Estimating the Lens Tolerance, we estimate and narrow the tolerances both for the physical parameters and for elements (such as matrices elements) in the mathematical model.

In addition, any time we use ray -tracing or the behavioral/mathematical model (matrices, formula, etc.) to mapping between image space→ object space (particularly for image space→ object space), we might not need to perform the same mapping for all points (e.g. all points on the edge). We might firs map a number of key points (major/minor axis endpoints, widest waist location, sharpest and bluntest/flattest points), then interpolation the location (like Z coordinate) of other points.

The essential methods presented in this application have all been disclosed in PCT /Π32013/056090 can be regarded as an elaboration thereof.

In PCT7IB2013/056090 §2.4 Image Simulation:

"There are other software capable of doing the same simulation. Code V is also widely used lens design software and gives reliable results. In addition, recent releases of PBRT (Physically Based Rending, from Matt Pharr & Greg Humphreys) also support tracing rays through real lens system. There is no essential difference if another software is used to do the same analysis." and when discussing our optical correction method, in §2.6 "Optical Correction":

"The online (dynamic) reconstructing method also works for the pendant, elongated droplet shapes shaped by to gravity and surface tension. We didn't use pendant shapes in illustrations above because spheres/circles were already enough for illustrating the problem (the variation of V according to D, R and T) and the all the inventive elements above applies equally well to pendant droplet shapes."

The optical lens (droplet volume measurement) optical correction methods broadly include two classes: a pre-stored database and online-computed values. This application combines them knowledge of the droplet position from accelerometer so the optical correction could work for tilted drip chambers.

Motivation

One of the main findings in 056090 application is that the measured droplet volume changes most significantly with its distance from the lens D, secondly with drip chamber's radius R, least with drip chamber thickness T. Their relationship is further shown in Fig.C.2-1 to Fig.C.2-3. From the lower sub-figure of Fig.C.2-1, and comparing the first and last plots in Fig. C.2-2, we see that for an offset D of 2mm, the measured droplet volume difference could be as much as 20%, therefore correction is needed. In clinical environment for example when the device is mounted on an IV stand, frequently it is tilted. Assuming that the length of droplet trajectory we measure is 20mm, and assume the maximum can be 7°, then the maximum possible D difference along the trajectory can be

20 Sin(7°) = 2.437 mm which according to the previous observation could cause over 20% difference in volume. Even if we standardize images on the optical axis which is halfway along the vertical trajectory, the difference in D with respect to the standard vertically center location is still larger than 1.2 mm, by looking-up in Fig.C.2-1 to Fig.C.2-3 we still see an around 10% volume difference. The fundamental reason for this is because lens designed to take do droplet volume photometry all need to be placed very close to them, and hence all subject to the shallow depth-of -field (DOF) problem.

It then seems obvious that volume correction is necessary particularly in aware of the tilting effect. In order utilize information in databases like Fig.C.2-1 to Fig.C.2-3, we need to know droplet's actual location when it appears in view of the camera.

Locating

There are two ways for obtaining the droplet's location:

1. Direct measurement (Please also refer to ^Summary and Flow Chart for discussion):

This is done by using one or more sensor's combination. For example, two cameras placed in orthogonal (for measuring Z axis depth directly; if using triangulation principle then no such principle) orientations as shown in Fig.2.6-2(1) of PCT/IB2013/056090 could be to find the Χ,Υ,Ζ coordinates (XY from one, YZ from another). Since when the droplet falls in oblique direction, the distances (D) to both camera sensors change which affects Χ,Υ,Ζ simultaneously, however the design of the optical system intrinsically enforces constraints on them, so the droplet's location can be calculated using both (XI, Yl) from one camera and (X2,Y2) from another, solving the proper equations.

Besides camera, ultrasonic wave, Doppler radar/sensor, infrared emitter(s)/receiver(s), other forms of electromagnetic wave can also be used measure droplet's exact location. They can be placed on orthogonal (to YZ plane) directions to the primary camera, in the YZ plane, or out of YZ plane and in oblique angle with it. They are not defined/restricted by their position, but by their function of measuring the droplet's location (primarily Z). For the radar/ultrasonic/infrared and various unmentioned types, they can work either by triangulation, or by sending a wave and measure the droplet's reflection (magnitude, phase, echo, etc.) time (times v, hence distance). The defining characteristic distinguishing them from traditional "drip counters" be that drop counters does not return depth/distance/location measurement, but here we use it as an important parameter for subsequent optical correction.

The two sensor methods discussed here essentially uses triangulation, can it can be achieved by single/integrated stereoscopic sensors (cameras, etc.), depth sensors (as in Kinect) and other integrated triangulation sensors which might also be count as one. Additional methods of using multiples are further described in §The Multi-Camera Approach later.

One can also use (ultra)sonic or any electromagnetic wave or laser (LIDAR, etc.) sensor, or even devices operating using RADAR principle, which measures the time between its outgoing and returning pulse to directly determine the distance (depth) of droplet in specified directions.

2. Calculation (using accelerometer, method I )

Using an accelerometer, the direction of tilting can be measured in real-time. The accuracy of orientation can be very high. For example, Freescale MMA8451Q has 12-digit readout for one g, each bit corresponding to 0.25mg.

The accelerometer can be calibrated at manufacturing time, or allowing software controlled re-calibration in the field. Because droplet' s speed is the lowest at its forming stage when it is still connected (pendant) from the drip mouth, this offers an advantageous image region to detect droplet's break-off time. The steps for detecting location would thenbe:

1. Taking forming stage (slow -changing) image (but because droplet not yet separated, these images are not well-suited for individual droplet's volume measurement), and detect break-off time.

2. When the next frame of image contains the droplet, calculate the time difference between frames.

3. Using g as acceleration, find the exact vector offset from the vicinity of dripping mouth. We can assume an origin point near the dripping mouth. Although because of the tilt, even at the forming stage the centroid of individual droplet is still inclined to some direction, the absolute offset of its centroid from the origin, particularly in the direction of the optical axis (the most sensitive variable D above), is small because the total "offset vector" is short and the tilting angle is small.

The method above used accelerometer to determine direction, camera to determine time instants, and the magnitude of acceleration is known a known constant.

A second method which uses accelerometer and camera to determine exact droplet location is illustrated in Fig.C.4-1. A droplet falling alone oblique vector is shown temporarily at location (x_trae=0, y_trae, z_trae) when its image was taken by the camera. In the case when we use only one camera (not the stereotypical/triangulation type in one assembly), the camera has only a 2D view of the droplet and don't know its exact depth in the Z direction. However, since Θ is known, if we could know the exact height of the droplet in the Y direction, then Z could also be inferred from that by timing

The droplet has corresponding image at (Ximage, YimageX and for different Z_trae, the magnification ratio is

y true different. For the lens we show in Fig.C.4 whose parameters and performance can be found in Fig.A.2.3-1 to Fig.A.2.3- 7 of PCT/IB2013/056090, for a square 11.4> 11.4mm² area centered at the optical axis and coincide with the mid-plane

y

of the drip chamber, the magnification ratio over the square XY area for different Z (the distance to the right

y true inner surface of the drip chamber changes between 5.5 to 8.5 in 0.5mm increments) is shown in Fig.C.4-2. The average value over each planar area (11.4mm by 11.4mm) is listed as:

- 0 . 3102531 - 0 . 3Q637S 3025 Si - S . 29S306 - Q . 255305 7S9

We notice that when the droplet gets close to the right inner side of the drip chamber, the magnification ratio increases, and decreases in the other direction (The 7mm distance is the location is the designed object location of the lens, please see §2.2 Cylindrical PVC Lens of the 056090 application). From the first and second (orthographic view) plot we see that at each Z location, the Y magnification ratio variation is relatively small and the plot appears rather flat. Normalizing the Y magnification ratio with respect to when the droplet is 7mm from the right inner surface of the drip chamber we obtain the lowest sub-figure, from which we see that the although the magnification ratio changes rather rapidly over the 3mm (8.5mm-5.5mm) distance ("rapid" because the 1.0384 ratio at 5.5mm, raising to cubic, causes 12% volume magnification change; and the 0.964 ratio at 8.5mm, raising to cubic, causes 10.4% volume magnification change), underscoring the shallow DOF problem we mentioned in ^Motivation, the change is however mostly linear.

Back to the question on how accurate we can estimate the (x^e, j , ^ztme) from (Ximage, image) :

1. the droplet's image (considering center only)'s Y coordinate at the image plane can very closely be approximated by y_tme · magnification(_ )_z=z (upper score denote average operation) because of the flatness of it the ratio.

the camera could only estimate (guess; by projecting, multiplication, etc.) the droplet's location at some particular plane(s), for example, the plane whose distance to the right inner surface of the drip chamber is 7mm, from now on called "7mm plane" for convenience.

magnification( )_z=¾

The estimated Y coordinate is now y_trm ^■—_{^^^^^^^^^^^^^}=- , the scaling factor, as shown in the magnification( )_z=7

3^rd sub-figure of Fig.C.4-2, is always between [96%, 104%], so the distance to y^ is smaller than

The true Z axis offset from the 7mm-plane is calculated by (lO— y_tru^Tan(ff) , and because of the estimate error in y_trm , the consequential estimation error here is bound by 4% · Ταη(θ)^■ y_trm . Θ in practice is almost always smaller than 10° (the Leaning Tower of Pisa leans at 3.99°), so

4% - Tan(\0°) = 0.705% .

A slope of the relative Y magnification ratio calculated from the 3^rd sub-figure (and data) of Fig.C.4-2 is 7.439%· mm^"1, and in practice the image is always captured at around the center of the optical axis (lowest distortion; also see §2.1 Blocking effects in the 056090 application), so we can give y_trm 's absolute value an upper bound of 3 mm. Taking these together, the resultant Z-dependent Y magnification ratio will have a maximum estimation error of

3mm · 0.705% · (7.439% · mnT¹ ) = 0.15% and its cubic (X and Y magnification ratio are very close) will contribute at most 0.473% error in volume measurement.

It is illustrative if we compare with the 0.473% error bound with the result without such correction: since we assumed y_trae of 3mm and Θ of 10°, the deviation in Z axis from the "7mm plane" could be

(10mm-3mm)xTan(10°)=1.234mm and according to the 7.439% mm^"1 slope we calculated above, this results in a 9.18% magnification change, whose cubic is over 27%, making the device totally unusable. This 20% plus estimate is also consistent with data in Fig.C.2-1 to Fig.C.2-3. The 0.473% (corrected) vs. 27% comparison shows how important magnification ratio correction is, and how effective accelerometer (or any device capable of measuring gravity's direction) can be of help.

Please also note that although in Fig.C.4-1 we show only droplet positions in the YZ plane, it is obvious from calculation steps above that the method also works when it is off the YZ plane. In addition, since at small Θ Tan[9] and Cos[9] are very close, Cos[9] can also be used instead of Tan[9] .

Yet another way of determining the droplet location is by calculating backwardly traced ray's intersection with the falling vector with one of its vertex at the dripping mouth location. This is discussed in §By backward ray-tracing section later (see Fig.C.4-3).

Location Falling direction

We explicitly note here that knowing one of {location, falling direction}, the other is simultaneously known. So any locating methods, including using the accelerometer per se, automatically serves both purposes.

The generalization of location and orientation detection

We summarize location and orientation detection methods more abstractly in Fig.C.7-1 :

First, there are in fact four principles for detection location (curve & line are used interchangeable below) :

(1) Surface (definition: two coordinates, or one surface equation/one indeterminate variable) + surface (two coordinates)→ point

Example: (XY + YZ) => two camera combine; camera + IR receiver grid;

(2) Surface + line:

If a surface location (such as XY coordinates from imager) is known from one meter, another line (such as reversely traced ray from imager/sensor) intersection the surface will allows us to solve all there coordinates.

(3) Line + line Example: triangulation

Trigonometry + line

Example: see our Tan[9] example above

Components Remitter (including light source; preferably directional, best to be collimated) and receiver)

Type: electromagnetism (camera, IR, UV), ultrasonic, laser, others There are several ways:

Fig.C.7-2 shows that at one end two rays transmitted by transmitter Tl and T2, vertically separated by dzl, transmitted parallel rays, and after reflection by the tranquil liquid, still in parallel, and hits the drip chamber surface, out of which two receiver (Rl, R2) cells detects the rays (the rays are preferably heterogeously decoded, as shown in §Using reflected wave section). The path within the drip chamber creates equal displacements and is not significant because the refraction index of PVC is 1.52, close to water's 1.33. A single pair, say Tl and Rl, we can only determine one plane (a point and a line projecting from Tl), and could not locate the exact incident point on the liquid surface. However, the distance between dz2≠ dzl, and this allow us to know the refracted ray (within liquid) 's direction vector's vertical component. From Tl to the line projecting from Rl there is only two such points that the connecting vector has this vertical component, and it is constrained by that the point must be within the drip chamber. From this we can definitely solve the incident point (if not sufficient, use other pairs). Hence, we obtain the liquid surface's surface orientation.

Particularly, it is seen that receiver arrays are preferably placed close to the drip chamber, best "wrapping" them to ease mathematical calculation. For transmitter rays the requirement is not that strict because after from air→chamber→back transmission, displacement of rays are small and close parallel rays are still mostly parallel.

For the liquid surface to be tranquil, the device (such as a pump) might intentionally stop the flow for a few second upon detection of shock, tap, movement (or suspect of that), which is not a feature of any previous IV delivery devices.

Another method is essentially equivalent to what is described in the next section §Using reflected wave: because refracted light through the droplet could also hit receiver, by uniquely identifying the transmitter location within a transmitter array (methods in § Using reflected wave), we instantly know the location of the droplet in one surface (facing the transmitter array) , or if using the delay information (phase, time, etc.), we instantly locate the exact 3D location of the droplet.

Another method is shown in Fig.C.7-3 : with any imager (electromagnetism, sound wave, etc.) the edge of the liquid surface can be obtained. Knowing only a plurality of the edge section points, the orientation of the emitter/receiver, and the fact that the edge is the intersection of a planar surface and a cylindrical inner wall (drip chamber), allows us to instantly calculate the surface normal direction (and hence drip chamber orientation). One grid of transmitter/emitter in principle allows to know the droplet's location equation with only one indeterminate variable (for example, depth unknown; surface coordinate XY at the dark area in the receiver blocked by the droplet), and two such pairs in principle allows us to accurately determine the location of the droplet.

Us reflected wave (including laser)

Component: at least one group of emitter(s) + receiver(s), electromagnetism or sound wave or laser or others.

This is illustrated in Fig.C.7-4: because the liquid is flat, the grid of emitter's collimated beam after reflection is still a collimated beam, and in theory only one ray will hit the receiver (depending on its size), and the receiver's detection doesn't require 90° incident angle. In order to know which emitter is the source of the received ray we have two methods:

(1) Homogenous emitter output, so we timely iterate emitters to emit rays subsequently and in high speed.

(2) Use modulation techniques to allow each, or each subgroup/subregion of emitters to emit wave of different frequency, strength, phase, etc. with either these two methods the receiver can know precisely the origin of the emitter. Then, the emitting ray and receiver uniquely determines a plane in which the liquid's normal lines, but its actual location yet indeterminate. To solve this we have two methods:

(1) Using another group of emitter/receiver, which might overlap the current group in either emitter or receiver but not both, to determine a second plane. The intersection of such two planes uniquely determines the liquid surface's orientation.

(2) For different possible incident/refracting point of the ray on the plane, the path from emitter to receiver is different. Therefore from the delay we uniquely knows which point is the incident point, and from incident and reflected rays' symmetry we instantly calculate the liquid surface orientation.

For locating droplet we use the same principle, but for one group of emitter we need to place more emitters at different locations around the drip chamber because the reflection due to the droplet chamber is scattered in many different direction. By either using the time-iterating emitter approach, or the simultaneous modulated heterogeneous emitter approach we could identify the emitter, therefore like for the transmitted wave case obtain an equation of the droplet having only one indeterminate variable. It would then be free to combine with others methods to obtain location of the droplet.

Using whatever methods, the orientation can also be calculated after determining two spatial (X, Y, Z) locations of the droplet and calculating their difference vector. Using Doppier Sensors (RADAR, laser, etc.)

In the first, Doppler sensors inherit all properties of transmitted (refracted) and reflected wave sensing above.

Additionally, a Doppler sensor placed, for example facing the YZ plane, could obtain the velocity's X component; another facing the XY plane obtains its Z velocity component; just as transmitted/reflected wave sensors can measure position hence speed, Doppler sensor can also do measure position in their respective viewing planes hence derive the Y direction velocity component. Or they can get it from camera measurement, etc. Hence, they provide another way of measuring the orientation.

This technique is similar to § The reverse fitting problem. Because from the image of the droplet (particularly its center) we could backwardly trace rays to restrict droplet' s true location along the ray, we could then at each possible locations along the ray (search at interval; recommend using more efficient search methods, including binary search, however optical considerations must be taken into account in determining searching strategies), parametrically construct a model of the droplet. This is essentially a more aggressive version of the techniques in § The reverse fitting problem, and goes by modeling the droplet and start the ray tracing directly over multiple possible locations along the line (backwardly traced ray from the sensor). The traced (from illuminator) result would then be compared with the droplet image, and for each droplet image performing this simulation might result in more than possible locations of the droplet, each corresponding to a unique falling orientation, and denote the set of possible orientations as U, ; performing this for a second image would further identify a new set of falling orientations U₂, and the true orientation could only be within U, Π U₂ . After a number of rounds we could then uniquely identify the falling orientation with high accuracy . The computational complexity associated with this technique has been discussed in §The reverse fitting problem, and in fact is completely tractable.

In additional to accelerometer, more traditional types of tilting sensors, such as all types of inclinometer, seismometer, tilt indicator, slope alert, slope gauge, gradient meter, gradiometer, level gauge, level meter, declinometer, and pitch & roll indicator, spirit level read by an electronic device (one modern type of tiltmeter), tilting sensor used in geodesy, seismic study, volcano studies, can of course be used. Orientation esi g Gyr sc

Any gyroscope has the property of maintaining their angular momentum regardless of the actual orientation of the object it is attached to (such as on aircraft within its own cage), and the relative orientation of it with the object can be sensed to detected orientation, whose use is ubiquitous in aviation and defense (as well as in the breakthrough Segway transporter). Therefore, we also propose the use of gyroscope (including MEMS type) in our system for orientation measurement.

Orientation ¾ssiag Compass (ma netometer)

We also describe the use of compass to detect orientation. The earth's magnetic field, although experiences reversal every hundreds of thousands of years, is relatively very stable during the life of an electronic product. At each different geographic location, its components have been measured and are available from databases like the World Magnetic Model. Fig.C.7-5 shows the particular example at Shanghai, China and we see that its vertical component has magnitude comparable to horizontal. Therefore, a compass (or a magnetometer) can be used to measure the respective components, and calculate the orientation instantly with high precision. The geographical location can either be fixed before shipping to particular customers, or allowed the user to set it or synchronize with external sources (PC, smart phone, etc.), or the device can further incorporate a GPS system which automatically gets the location.

Orleatailos sks a weight/pressure asymmetr

As shown in Fig.C.7-7, the drip chamber can be supported by multiple supporters (arms, etc.) at the lower side, or hanged from (suspended, etc.) equally from the upper end (possible with small hooks made on or attached to the drip chamber). When upright the supporters are expected to measure same weight because of symmetry, but when tilting the sloping water surface will cause uneven pressure on different supporters, whose measurement allows us to calculate the orientation instantly. Scales (ft) of O.OOlg accuracy and digital output for measuring precious metals are available only a few dollars which makes it an economical and viable approach.

Orientation aslog Physical property asymmetry measurement

As in §Using transmitted (in general refracted) wave, a grid of Physical properly sensor probes can be placed to wrap (at least partially) the lower end of the drip chamber. The physical property of drip chamber plastic (particularly thermal: heat capacity, heat conduction rate, etc.) differs significantly with that of water, so when the drip chamber is upright the probes (preferably temperature sensors, and preferably there are also grid of heat generating sources placed together with the probes) senses symmetric properties (temperature), but when the liquid surface tilts certain we could detect on highest point and one lowest (due to temperature distribution, etc.; other physical/thermal change permissible) In the case of temperature distribution, we can use thermal simulation techniques in PCT/IB2013/056090 to pre- calculate distributions at different ambient temperature, liquid surface height, liquid kind (drug), and use that to enhance the real-time detection. From liquid surface high and low points and intermediate points (edge, etc.) the calculation of tilting is also instant.

OHen tatton/Loca tkm n ig ll n locat on

Using any of the location method discussed throughout this application (imager (generalized), wave, etc. all), even if we do not identify the droplet's location at the same instant when it is captured by the volume measuring camera(s), as long as we could locate other locations of the droplet in its trajectory, we could also calculate the falling orientation from the it and the dripping mouth location.

When droplet impinges the liquid surface, there is such a clear sound which could be heard meters away by ear. Because sound is air's mechanical wave, and is caused ultimately by vibration of the surface, we could place an array of vibration detectorS (sound wave, or any detector capable of detecting mechanical wave/air vibration; in concrete terms microphone, etc.) wrapping or placing around the drip chamber, then

1) from the time/phase difference each detector detects the impinging, we know distance difference because mechanical wave (including sound, or vibration, or material of the drip chamber, or material of the device) 's speed in medium is known (liquid, air, or material of the drip chamber, or material of the device) is known, we could infer the impinging location, and hence falling direction and device orientation.

2) from the strength/magnitude difference whose behavior could be calculated from the wave's differential equations (or experiment determination), we could also inversely solve for the impinging location.

It is further proposed that for any sound detector, because sound speed in air is 343m/s, we proposed the detector(s) no farther than {10mm, 20mm, 40mm, 60mm, for limiting claim} away from the liquid surface (limit to 1/3 height from bottom of the drip chamber), and correspondingly sampling interval shorter than 2mm(for differentiating locations)/(343ms^_1)=0.00583ms (or 0.0116, or 0.0233, 0.0349ms, for limiting claim).

Passive Sesssisig

Exam le I: llsisig Capacitance Sesssksg

This is one example in which the sensing device does not actively dissipate energy. Consider putting two capacitor plates, divided into a grid of small cells, parallel with YZ plane of the drip chamber, after charging an electrostatic field is established. The capacitance between each cell pairs is determined by relative permittivity ε_Γ between them, of which PVC and air are known and is pre-calculated. Water however, have high ε_Γ of 80 at 20°C, and a 0.05mL droplet has diameter 4.57mm which is almost 33% of the inner diameter of a typical drip chamber. Such a high permittivity compared with air's 1, significantly changes the capacitance between cell pairs which it is most closely to d which in turn causes Q, and current to change. Therefore we easily detect the location of the droplet in the capacitor array's viewing plane.

In addition, just as we have shown for imagers sensors, this disturbance principle of capacitance change can also be used to detect edge of a tranquil tilting liquid surface, from which we instantly calculates the orientation.

Similar idea is exploited, by putting magnetometer arrays parallel YZ plane (for detecting Z; others also allowed). Water is classified as diamagnetism, however its dimension, as above calculated, equals 1/3 of the drip chamber diameter, which would guarantee an appreciable disturbance of the magnetic field between around its nearest magnetometer cell. This signal, after amplified by amplifier (what all Hall Effect sensor used to magnify their extremely weak signal level), can be used to locate the droplet's position in the viewing plane. Modern magnetmeter have been reported to have sensitivity reaching the orders of μΤ, fT or aT, and would certainly be able to detect the passage of droplet which relatively is very large in its sensing space.

Supplementary methods

Additionally, orientation can also be calculated from visual indicators, which include pendulum (preferably with ticks), bubble (liquid/gas) orientation meter, and the image could read from at least two such meters to determine the location.

Alternatively, the device might be electronically coupled (including wireless) with orientation metes attached to the IV post, monitoring gadgets attached externally to the device itself, to send images to external computers to ask that to calculate orientation, and to receive orientation monitoring from external cameras, and so on.

Alternatively, there might also be visual indicators (such as dial, or display) that can be viewed by one or more imagers so that the orientation is conveyed to the device by analyzing the acquired image. Alternatively, the drip chamber might also have pendulum/spirit level or other physical orientation indicator affixed to it, either outside the chamber or within it, so that the imager might read from their positions/gestures/postures and infer the orientation. Correction

The following correction method only requires droplet' s location (which implicitly includes orientation since orientation can be determined from the vector between breaking-off location at the vicinity of the drip chamber and the location when the image is being captured) as its input, therefore works with any of the droplet locating methods above, and could also work with other methods not mentioned here.

Using essential the same method as Zemax programming simulation which varied {D,R,T} above, we might further include in four (at least) additional variables:

These four variables determine location of the droplet in the space with respect to the dripping chamber, and by varying one or several of them we obtain a conical (we can also of course simulate cubic grids of droplet locations, but conical corresponds to the physical reality) arrangement of droplet instances, like showing in Fig.C.3-1.

A standard 20 drips/min IV set's droplet has radius = 2.28539 mm

4π / 3 when the shape is perfect spherical, but in most time of the falling it is under the air resistance and gravity and the shape deviates from spherical (see Fig.C.3-2 and Fig.C.3-3), so it is necessary to use ellipsoid, parametric forms or other representation to model all the different shapes.

We then generate a "location-correction library" using the following steps:

1. Looping for {volume, e, h, θ, τ} combinations.

2. For each three-dimensional droplet generated, projecting light to illuminate it and simulate its image in the camera sensor's plane. This can be done either using self -written code (please refer to §2.2 and §2.6 of 056090 application, where we used C code (compiled to DLL) to model drip chamber's cylindrical surface lenses), or using software packages capable of doing 3D photorealistic simulation (Matt Pharr's PBRT, etc.). 3. Analyze the images to obtain the relationship between the measured volume from the image and the actual volume. From these statistics, we can find the which of the {e, h, θ, τ} is the principal influencer and which ones are lesser, just as we have found that (by decreasing influence on measured volume)

D > R > T in PCT/IB2013/056090.

4. The relationship can either be stored in a database (like a large dimensional (more than 3D) database), or from we can find interpolated (by statistical methods, by machine learning methods, etc.) formulas (mapping/transformation matrices, etc.) to represent them.

5. {e, h, θ, τ} can optionally to combine with {D,R,T} and volume to form more comprehensive correction database (formula).

6. When the device is in use and is tilted, obtain droplet's location using methods in the previous sections and in section ^Summary and Flow Chart below, and the database or formula in this section, to approximate the "true" droplet volume. As having been shown in Fig.C.2-2, D -difference of 2mm could have as large as 20% influence on volume, and it is reasonable to expect similar results from h, τ or so on. The improvement of volume accuracy due to the correction would be very significant.

Please note: in Fig.C.3-1, the simulation library was constructed from conical arrangement of locations, and the orientation of droplets (pointed by their axis of rotation direction) always coincides with the falling vector, and the vertex of the cone is at the dripping mouth location. These arrangements correspond best to the reality. Of course we could also simulate in cubic spatial grid (which entails more locations droplet instances), and for each grid multiple orientation of droplets. The essential of the correction is always the same as to construct a pre-stored {location, shape, volume}→correction library which the devices uses during monitoring.

Calibrating method (1): By robotic arm / real objects

An alternative way to generating the simulated "library" data is to use real objects. An robotic arm positioner can achieve any of the {h, θ, τ} combinations, and by attaching precision-made metallic (or other material) droplet-shaped objects to it, we can achieve the same effect that the device can take images corresponding to different {e, h, θ, τ} and volume combinations. The image and database processing would then be done on the device per se instead of offline on a typical more powerful computer. An advantage of this method is that data obtained this way is more realistic because lenses are not exactly the same due to manufacturing and assembly tolerances, and doing this "real droplet" simulation automatically compensates for all of the tolerances. This is illustrated in the upper part of Fig.C .2-4 in which a robotic arm positioner moves a sphere(s) (or ellipsoid, or of other types of droplet profile) of varying droplet dimensions (20 drip/min, 15 drips/min, etc.) inside a drip chamber, and after placing the object at each exact position, the robotic system sends the device a signal (or not, because the device can recognize itself) informing it the exact position of the object, and the device would start the calibrating process, eventually obtaining a database library or formula (matrices, etc.) as described above. Please not that although in Fig.C.2-4 we show a cubic grids of positions, because of the falling trajectory sweeps a cone, we can of course also use conical positioning. And the robotic arm could attach to its moving head objects of different shape (ellipsoids with different e, etc.) and volume.

(The following paragraph describes using real drip chamber material or material with similar index of refraction, placing in front of the calibrating objects to better simulate the effect of PVC drip chamber lensing, and claims corresponding priority in PCT/CN2013/086477).

To better simulate the setting of a droplet behind the PVC chamber (effectively forming a lens, see section 2.2 (§ A.2.2) in PCT/IB2013/056090 "Designs for a very precise IV monitoring system based on computer vision technology"), we could (not imperative) put an actual PVC chamber (see Fig.C.4) in front of the calibrator. The PVC chamber is preferably made of PVC with a low degree of plasticizer so that it is not flexible and has a definite shape (curvature) or even other types of plastic/glass as long as its optical properties (index of refraction, Abbe number, partial dispersion) match or are close. For example, quartz has index of refraction 1.544296 which is very close that of PVC drip chamber material, and they can be manufactured to the thickness (0.5mm) similar to those of drip chamber with high precision.

Calibrating method (2): Z-Shifting Calibrator

Another way of using real rather than simulated object for to obtain the correction is to record the images of a calibrator shifting in the Z direction (optical axis). This is shown in Fig.C.2-4 in which inside the same PVC drip chamber (or from material with close index of refraction; or without; see above), we let a precision-made calibrator moves along the optical axis, and the camera captures successive images, associating with each its exact location in the Z direction. Openings in the calibrators in the drawing showing lines which permits backlights to pass, and they can also be grids of dots, other types of patterns (grids of circles/ellipsoids, resembling upper parts of Fig.C.2-4, or grids of concentric circles/ellipsoids to better simulate droplets of different dimensions, etc.).

By "calibrator", we broadly refer to any objects of known size, such as rulers, rigid objects of specified dimensions or having notches separated at fixed distances, so that they could be used to establish pixel-distance relationship. It is preferably be reticles since they can be made to accuracies of within microns. And by the use of these calibrators, the device would know at Z plane and at each of the specific Z-plane's X-Y location (XYZ follow optics convention) the pixel→distance mapping relationship.

Please note that as the reticle moves along the Z-axis, we can also measure for each source location, the blur it caused due to defocusing. This is discussed in § Difference with edge blur and edge identification. An enhanced version in which rotational directional lights are placed after the reticle "light sifts" is shown in Fig.C.4-7 which can also simulate for each 3D spatial source location, different directions of emitting light's response on the image sensor plane, including blurring due to defocus.

Both the simulation and calibration methods involve processing (condensing into compact form, etc.) of the simulation results/calibration data, and these can be done either outside or within the device's own processor. By integrating the processing ability into the processor, the device acquire the ability for doing calibration after the device has been used for some time and some of the optical characteristics has changed (such as due to shaking, etc.) and needs updating these data.

Estimating the Lens Tolerance

In ^Summary and Flow Chart below we mentioned "estimate various crucial manufacturing tolerances", and we give the details here. The typical (not meant to be restrictive) type of tolerances encountered in optical manufacturing and assembly are:

Assembly MaaufactBring

Tilt Radius (millimeter, Fringe)

Aspheric coefficients

Thickness Thickness

Including that of sensor cover glass and air

gap between cover glass and sensor plane

Decenter

Zern irregularity

spherical and astigmatism irregularity

Glass Index n

Glass Abbe number

In addition, as shown in Fig.C.2-5, after the optical lens is put into the system with the drip chamber, (as having been shown in PCT7IB2013/056090, drip chamber is also part of the overall lens system) the distance, tilting, decenter, radius and thickness, etc., might also have tolerances. For the variation on drip chamber radius (R) and thickness (T) we have already given descriptions in great detail in this and previous applications on their modeling, simulation and correction. The ensuing discussion elaborates the detailed general framework for the estimation of all (important) tolerances.

Please refer to Fig.C.2-6:

First, in the manufacturing and assembly process of the lens we would try to measure with best effort the various parameters. In this process lens out of our designing tolerance specification will be rejected, and the parameters of lens passed the test are recorded.

Second, there is an iterative process (might be as short as only one iteration, depending on lens manufacturing and assembly quality and device accuracy requirement) of narrowing tolerances after which we obtain more "true" lens parameters corresponding to what is installed on the system. Within each iteration:

1) A unique input is given. The input is meant to elicit particular response(s) of the lens system so we could inversely narrow tolerance range.

a. The input might be a set of calibrating object(s), which might be reticle, any object with

marked/notched/ticked/precision-made fixed-distance features (points, rectilinear lines, etc.), checkerboard pattern, drawing (image), circular or other pattern, MTF testing pattern, etc. By unique, we mean that even for the same set of calibrating object(s), if its orientation (different tilting as illustrated in Fig.C.2-5), center location, location in the Z axis, illumination, the responding imager's region of interest (which area the imager analyzes) or any associated variable changed, it is regarded as a different input. This "uniqueness" definition also applies to other types of inputs.

b. The input might also mean that we use a certain type of optical instrument to measure its particular set of parametric. For example, the MELOS 530 measuring combination from MOLLER-WEDEL OPTICAL GmbH can be used to measure at least effective focal length, back focal length and radii of a lens system which also fits under our above definition of "elicit particular response(s)" of the input. The input might also be a collimated beam, an interferometer (of all types); Foucault, Wire, and Phase Modulation Tests; Ronchi Test; Hartmann, Hartmann-Shack, and Other Screen Tests; Star Tests; Aspheric Wavefronts and Surfaces tests; Zernike Polynomial and Wavefront Fitting; Phase Shifting Interferometry; Surface Profilers, Multiple Wavelength, and White Light Intereferometry; MTF test; Optical Metrology of Diffuse Surfaces; Angle, Prisms, Curvature, and Focal Length Measurements.

2) For each input the output is analyzed (either algorithmically or with human intervention; if the camera gets the input, the device might have a signal coupling with an external computer (workstation) so that the data can be uploaded to the (presumably) faster computer to facilitate analysis) and then download the analysis result (including the narrowed tolerances and updated "true" lens parameters). The data is optionally stored, particularly if reticle (or any regular structure, preferably rectilinear or circular, is used) is used and shifting its Z direction so the 3D grids can be used to for later convenience (such as interpolating continuous pixel→ spatial grid mapping; see ^Calibrating method (2): Z-Shifting Calibrator).

3) If the information gathered from the output response is enough to reduce the range of any variable (the indeterminate variables are preferably ranked in importance, this can be done using our simulation analysis technique discussed in 056090 application which for example identified D>R>T), the inversely narrow their (its) range in the tolerance space (not all tolerance space points (high dimensional) could yield the same/close output response; this narrowing can be done by searching the space, using pre-calculated relationship, solving equations, statistical analysis, machine learning, etc.). If the output response gathered is not yet enough, not well enough for various reasons not suitable/opportune to narrow the particular tolerance variable(s), store it optionally for later iterations' possible use, and move to the next iteration which feeds a unique new set of input.

4) If after a number of (zero if the lens is perfectly made and assembly; possibly one is the lens making an assembly quality is superb and the principal tolerance influence is immediately identified) iterations the tolerance are narrow enough to approximate a relatively very "true" lens model, the stop the iteration and update the lens parameters with estimated values on the particular device. Therefore, the true pixel→ spatial point mapping can be calculated from the updated "true" closest lens model, and can optionally make use of the Z-shifted reticle data described in step 2) and ^Calibrating method (2): Z-Shifting Calibrator.

We particularly note that this tolerance narrowing/estimation applies to all phases of lens manufacturing/assembly. This include when the lens is assembled per se without camera sensor, when it is assembled and aligned with the sensor, when the lens-sensor combination is mounted into the device, or the lens if first mounted into the device and then we attach, align and adjust the sensor to the lens.

It is also proposed that, because in hospital use this precision optical device might experience shock, lens contamination and for various reasons might need maintenance so that parts (lens, etc .) might be replaced and reassembled in hospital, the device would have either or all of the following features:

1) Incorporating at least partially the tolerance narrowing/estimation as mentioned above feature into its own software so it could update to the "true" lens parameters automatically.

2) Having a signal interface (by a communication protocol, hardware interface, for example USB, I2C, SPI, Ethernet, wireless) to an external field computing device, so that it can optionally upload the data to the said device, which performs the same task as the aforementioned "workstation" (or computer) does, and later receive analysis results, narrowed tolerances, and updated lens parameters.

These two features would make the device field-adjustable without the need to return to factory for maintenance.

By backward ray-tracing

This has been discussed in detail in §2.6 Optical correction of PCT/IB2013/056090. For each unique image captured, because we can know {h, θ, τ} from accelerometer or various other means (see §Flow Chart and Summary), we could determine droplet's actual location. Then, doing a reverse ray-tracing from droplet images backward to the droplet's plane, we reconstruct the original droplet symmetric profile, and from that calculate the true volume.

Determining droplet location (using accelerometer, method II)

The intersection of backward traced ray can be used to determine the location of the droplet, which does not require capturing the "break-off image and register the exact time when it separated the dripping mouth. This is illustrated in Fig.C.4-3 : For each droplet image we calculated, we first compute its center's location (or other locations, but preferably those with special geometric meaning such as major and minor axis endpoints, see the discussion on Shape Analysis later) (usually the centroid) Pixel_center, the do backward tracing from the center pixel until the ray enters the spatial region within the drip chamber. In theory, the droplet's center (Physical_center) can coincide with any point on the line segment within the drip chamber for its image to be at the Pixel _center, but only one these possible locations lies on the trajectory of falling (vector), which can be calculated from accelerometer data. In implementation we might calculate the intersection of the reverse ray with the falling vector beam, instead of a single line, or find the projection point (on the connecting line segment perpendicular to both) of the ray onto the falling vector direction, and use that as the droplet's center location.

The orientation of the droplet's "plane" is not necessarily perpendicular to the optical axis; we recommend using a plane coincident with the point (denoted as Pj) at the droplet's location, and perpendicular to the line connecting P_l and the center of the entrance pupil. This works best when the shape of the droplet is spherical or close to spherical (see Fig.C.4-3 illustration). For elongated or compressed (bean-shaped, by air resistance) droplets whose axis of rotational symmetry does not lie on the said plane, some correction (from the database, formula, etc. described above) might be needed. Other orientations can of course be used, with the understanding that perpendicularity to optical axis is not always the case, especially for falling vector directions which does not intersect the optical axis (off the YZ plane (conventional in optics, Y is the up direction, Z to the right)). In general, the task of reverse ray -tracing when the gravity vector is off the YZ plane is not trivial.

Alternatively, as we have elaborated in the 056090 application, we can compute reverse point-spread functions (PSF) for different {(optionally include e), h, θ, τ} U {D, R, T} combinations and use reverse convolution to reconstruct true droplet profile.

Rotation

Images of tilted droplets are in general not in the upright position, therefore when determining their volume using the "summing up stacks" method, the orientation of its rotational axis needs to be determined. This can be done from the image itself if it assumes an ellipsoidal shape by first finding its major (a) and minor axis (b); another way is to use the location information calculated from the accelerator data, then calculate the location point's image location (forward ray-tracing), the determine orientation from image point's relation to the center of image (on the optical axis). Then we either first rotate the droplet image then integrate slice by slice, or integrating directing the sloping/leaning slices in the original image.

Simulation Examples

An example in Fig.C.5 shows one droplet of the same 50mm volume, but with different {e, h Θ, r} parameters hence at different location, and the respective simulated images. The second row, first column image shows a tilting (in the image, not object space) several degrees off the upright direction, and the 2^nd row 2^nd column image of an elongated shape. It can be seen from these results that corrections are needed to normalize them to yield true droplet volume measurement.

Summary and Flow Chart

In this ending section we summarize the rationale and methods of optical correction. As having been shown in Fig.C.2-1, droplet volume measurement result is particularly sensitive to D, the distance offset from the intended object plane, and a 2mm offset can cause over 20% volume differences. This shallow depth-of- field problem is an intrinsic property of lenses forming images of near objects. The microscopic pixel-to-physical mapping at different XY planes offsetting in the Z direction from the intended object plane changes sharply with increasing Z offset value. The fact that IV stands have tilting angles makes this problem unavoidable, and necessitates methods for correction.

The prerequisite before any correction can be made is first to know the actual Χ,Υ,Ζ location of the droplet image, and this requires at least two sensors. Any two sensors which can both capture the relative position of droplets in their field of view (include image sensors, IR sensor arrays, small ultrasonic sensor arrays, etc.) can be placed at different locations (preferable orthogonal) around the drip chamber and we could then use locations obtained from the two sensors (two receiver plane) to solve for the Χ,Υ,Ζ location. Alternatively, an accelerometers could be used to measure falling vector direction, and by combining this with the initial droplet "break-off location and moment information (obtained from camera, IR transmitter, receiver, ultrasonic sensor, etc.), or by reverse ray -tracing from droplet image and finding the ray's intersection with the vector direction, we could also find the Χ,Υ,Ζ position of the droplet.

Please note that by using "at least two sensors" we implicitly refers to conventional sensors (optionally plus lens) which in the ideal case has one-to-one corresponding between object space point and image space points, and the number two means we use the triangulation methods for determining the droplet location. The triangulation also applies to the case when a vector direction from accelerometer is intersected with reversely traced rays.

The correction methods are then classified according to their susceptibility to several factors, as shown in Fig.C.6-1.

If the lens is made to perfect agreement with the lens design formula, then the lens model itself provides all the information needed to reconstruct the object from the image. After locating Χ,Υ,Ζ distance of the droplet, we could reversely trace backward rays until it intercepts the object's plane. This works well when the image edge is sharp and when its trajectory is in the YZ plane, but the performance degrades when image is blurred, and when the falling trajectory is off the YZ plane, locating intersection plane/points is also non-trivial. It is also affected by manufacturing tolerances. The droplet volume is determined from the intercepted profile in the object space, using methods such as slice-by-slice integration. Additionally, we might avoid the burden of reversely tracing rays each time by pre- computing from the lens model a pixel-distance mapping relationship, stored in database library or formula (matrices, etc.) for different (Χ,Υ,Ζ) points, and retrieve that at running time. In this way, Pixel-wise integration in the image space, with each summing element multiplied by the corresponding pixel-distance factor at (Χ,Υ,Ζ) location, is equivalent to physical distance integration in the object space. This is shown in Fig.C.6-2.

One way to compensate for the manufacturing tolerances is to calibrate using one reticle location, or the images of a reticle shifting along the Z axis from the intended object plane. If only one reticle location is used, accurate ipixel- obiect locationl relationship can be obtained for the plane at the said location, and we can therefore compare the relationship with the lens model to estimate various crucial manufacturing tolerances (see ^Estimating the Lens Tolerance), and from that infer the pixel-location relationship on other XY planes. If multiple reticle locations are used, particularly when a dense array of grids is sampled, we can directly interpolate the [pixel-object location] at any specific object space location. The volume could then be determined by directly mapping from pixels to object space locations, and determine the volume from the object space droplet profile.

The two previous methods integrate in the object space, and could be classified as the "microscopic" ways. Another class of methods uses a macroscopic scaling factor in the last step.

First (can be deferred) and most reasonably (not imperative, but it is the mathematically most correct way) the sensor image is rotated. Then there are two choices:

1. (1) We calculate the volume (preferably slice-by-slice integration) in image space under pixel units, and directly use Χ,Υ,Ζ information to retrieve a scaling factor from either the simulated or real volume correction database / formula( matrices), and directly multiples this macroscopic factor to get the droplet volume. (2) or we first map the pixel to a reference plane (optionally reticle-calibrated), and retrieve from either the simulated or real volume correction database / formula( matrices), a correction (scaling) factor between the object location on the said plane (at specific X, Y location) and the true X, Y, Z location, and do the scaling. This looks more like to a two-step method in which the first step behaves "microscopically" in its pixel-wise integration, and in its second step "macroscopically" uses the scaling factor.

2. Without doing the integration, we directly extract from the image important features (eccentricity and major/minor axis length for ellipsoids, important coefficients for other parametric models, circumference, pixel count (not recommended), etc.), and from these information directly identifies the closest model (in the simulation, droplets at each location can have different volume and shape, particularly for computed simulation but also for robotic arm positioners) in the simulation/object calibration (robotic arm positioner) database at the Χ,Υ,Ζ location. In this way, we circumvent the intermediate integration steps and works solely in the "macroscopic" direction.

The reverse fitting problem

All methods discussed above falls into the larger scope of object space— image space and within its domain we see the existence of both microscopic (pixel→physical distance mapping and integration) and macroscopic (final scaling) methods.

If we look out of the domain of mapping from image space to object space, after some thought we realize that the solutions above has its dual in the reverse direction, i.e., mapping from image space to object space.

We know that all droplets have smooth profile, and its shape is governed by the dynamics involving gravity, air compression pressure and the fluid mechanics within its own space. Besides, their shape is rotationally symmetric. Therefore, we might represent their 2D profile using parametric 2D curves or directly parametric surfaces/bodies. There has been extensive study on this particularly in metrology, for example, in Pruppacher & Pitter, UCLA 1971, - A Semi-Empirical Determination of The Shape of Cloud And Rain Drops

Beard & Chuang, UIUC 1987 - A New Model for the Equilibrium Shape of Raindrops

Cosine series has been shown to closely match the shape of true raindrop profile (please refer to the papers for the exact formula). Therefore, we might prefer to use such a series or other parametric curves/surfaces/bodies with strong "fitting power" (a list without intention to be restrictive: ellipsis/ellipsoid, Fourier (and any generalized Fourier Series over orthogonal basis, and extend the generalized orthogonal polynomials below), Laplace transform and particularly its inverse, Z-transform and particularly its inverse representation, Bessel, spline, cubic spline, B-spline, Coons Surface patch, Gorden surface, , NURBS (which is a generalization of Bezier and spline; it is so widely used in representing huge varieties of geometries), Sinusoidal series, power series, polynomial), interpolated functions,using Laplace-Young and/or Naiver/Stokes equation which is more pertinent to fluid dynamics, etc.) to represent the original shape of the droplet. For naturally realistic smooth curves/surfaces/bodies mathematically there are virtually infinite choices to approximate/represent them, either using elementary functions or even artificially constructed/contrived functions, therefore it is certainly impossible to exhaust the list. However, we believe the proposition of using parametric representation of IV droplets in its optical measurement is new and the number of examples given above should justify the coverage of the whole class.

In addition, power/Taylor series expansion applied to/associated with any of the parametric representations above could also be used.

Particularly, mathematical functions with orthogonality have the desirable property that the coefficient terms or other parameters can usually be determined individually by taking the generalized inner product (e.g. integration; can have a weighting function w(x)). This makes them an ideal choice as fitting curve candidates. We therefore recommended the generalized orthogonal functions (including Fourier class functions / generalized Fourier class series) which in a hierarchical listing include: olutions to

Jacobi polynomials

Gegenbauer polynomials

Laguerre polynomials

Hermite polynomials, Hermite interpolation

Chebyshev polynomials

Legendre polynomials Wilson polynomials, which generalize the Jacobi polynomials Meixner-Pollaczek polynomials continuous Hahn polynomials continuous dual Hahn polynomials

Askey-Wilson polynomials

Discrete orthogonal polynomials

Racah polynomials

Hahn polynomials

dual Hahn polynomials

Meixner polynomials

Krawtchouk polynomials

Charlier polynomials

Sieved orthogonal polynomials

sieved ultraspherical polynomials

sieved Jacobi polynomials

sieved Pollaczek polynomials

orthogonal polynomials on the unit circle

Rogers-Szego polynomials

Zernike polynomials

Multivariate orthogonal polynomials

Jack polynomials

Hall-Littlewood polynomials

Heckman-Opdam polynomials Koornwinder polynomials

Askey-Wilson polynomials and more generally include any suitable polynomial included in the Askey scheme [Andrews, George E.; Askey, Richard (1985). "Classical orthogonal polynomials".], and also include Biorthogonal polynomials, which is a further generalization to orthogonal polynomials.

The fitting/interpolation methods might also include

Linear interpolation

Bilinear interpolation, Bicubic interpolation, tricubic interpolation Nearest neighbor interpolation Cubic (Hermite) spline interpolation Catmull-Rom / Centripetal Catmull-Rom spline

Piecewise cubic Hermite interpolation (PCHIP) which preserves monotonicity and the shape of the data

Biharmonic fitting (one of MATLAB's interpolant methods)

In addition, the representation might also include (might partially over all "point set") piecewise functions, particularly piecewise linear functions, over different ranges along the profile viewed by (each) camera(s) or just different subsets of the object space point set. Using piecewise linear functions can result in non-smooth joints (tangent unequal, etc.), but as long as the linear sections approximates the shape well, we can still achieve high accuracy volume calculation by summing the revolution (not necessarily always 360°) corresponding to different linear sections.

It is also proposed that at any step in the fitting when least square method (e.g. regression) is used, the LOESS and LOWES S (locally weighted scatterplot smoothing) might be used to overcome deficiencies of linear fitting.

In order to determine the parameters/coefficients of the parametric curve, follow the following steps:

1. (Optional) Identify the Χ,Υ,Ζ location of the object space droplet (or loop for all possible locations within the drip chamber). The orientation of the rotational axis must in close (allow tolerances due to measurements, calculations, etc.) parallel to the drip's falling direction determined either directly from accelerometer of from droplet location Χ,Υ,Ζ.

2. Loop the coefficients. Generally lower order terms control the general trend of the curve while a higher order term has more influences on sharp local transitions.

3. Tracing rays from illumination sources to create droplet images in the sensor plane. 4. Compare the simulated image profile (particularly edge) with the original image, if the deviation is large, goto step 2; otherwise converge.

The above illustrates iterative methods; we could also explicitly solve the coefficients for certain parametric forms. For example, if a 20th order polynomial is used to fit 20 key points of the droplet profile in the image plane, solving the equation

Po

A Y

10

Involve only O(20²)=400 operations, which require virtually no time for DSP processors.

Please note that we have explicitly defined the scope of our parametric (not limited to series, for example, if modeling droplets using ellipsoid equation then infinite series is not required at all) representation to include both curve, surface and body cases.

Computation for ray-tracing (object to image):

The lens in Fig.C.4 has six surfaces. For spherical lens shapes, finding the incident point of a specified ray starting from S and in direction d={d_x, d_y, dj to s spherical surface center at O with radius r involves first solving the quadratic equation:

(S_x + td_x - O_x f + (S_y + td_y - O_y )² + (S_{z +} td₂ - 0₂ )² = r² to find t, followed by a vector rotation with ΑΘ calculated from the Snell equation. Even if there are as many as 20 surfaces, the operations above is still very efficient on modern processors.

Computation increase due to the larger number of rays from illumination source

This is in fact not necessary because of if we look at Fig.C.3-2 and Fig.C.3-3, the dark edges were created by strong Fresnel reflection for incident backlighting at large incident angles. Since we need only to fit edges, we only need to project light from a "patch" of very small area on the entire illumination space, the area (defined as the actually ray emitting area. This is particularly pertinent concerning that we don't need to simulate rays that will pass through the droplet's central area, see discussion below) of which in general does not need to be larger than two times (we list some other alternatives for possible claim restriction: <3 times, or <5 times, or <10 times, or <=15 times, or <20 times, or <25 times, or <30 times, or <50 times) the parametric 3D shape's area of projection on the illumination background (in most cases one uses uniform backlight rather than spot lights so we could speak of "backlight"). In addition, the inner area corresponding to the within the edge of the droplet shape does not need to emit light since much of its light passes through the droplet due to small incident angles (Fresnel) and doesn't contribute to dark edges.

More specifically we give several independent limitations:

1. The area, ray number distribution over the area, ray direction distribution over the area, and the relative low threshold value at which the ray tracing is terminated, varies with different droplet locations. This applies equally to either a fixed illumination and ray source model or Monte Carlo ray tracing.

2. The area of the illumination ray source is no more than 80% of the actual physical ray source area in the device.

3. The area limitations corresponding to the list figures given in the "patch" discussion paragraph above.

Processor Considerations

The highly optimized fitting procedure would cause no problem on modern processors. First, many smart phone processors already run at higher than 1.4GHz and have as many as four cores; secondly, ray -tracing is inherent parallel and could be run on GPUs (what GPU originally designed for, and smart phones like iPhone already include mobile GPU). Third, recent progress in Quantum Anomalous Hall Effect show promising potential in building processors with low-heat dissipation and hence dramatic increase in speed (Science 12 April 2013: Vol. 340 no. 6129 pp. 167-170).

The volume of droplets, after all parametric coefficients have been determined, would then be calculated directly from the coefficients, instead of pixel- wise integration (although we do not exclude this).

The forward fitting problem

Another class for parametric model fitting, which does not require tracing a multitude of rays from the light source to the image plane, is by fitting to a particular set of points. By "set of points", we include discrete points, connected points which forms part of a curve (including droplet profile after being intersected by any plane (without restricting its orientation to be perpendicular to Z axis)), and in a more generalized sense, any grids in the 3D spatial location. The difference between this method 1) integrating then scaling or 2) integrating with corrected pixel→ physical distance relationship, is that the droplet volume is calculated from a model fitted to the "set of points" above. This is done by

1. Mapping a set of points from the image plane to object space.

1. These points preferably include geometrically more significantly ones, particularly major and minor axis endpoints is assuming ellipsoid profile or if the image does resembles ellipsis, endpoints of symmetric axis (if image differs from ellipsis), center points, points evenly separated along the edge (e.g. four, eight, widest waist location, sharpest and bluntest/flattest points).

2. Their locations in the object space is obtained backward ray tracing, or through reversely mapping matrices between image and object space.

2. The object space points are then assumed to be on the surface of, or within, the spatial scope of the true droplet. The droplet is then assumed to be of a parametric form with all the possible representations and parameters discussed extensively in §The reverse fitting problem above. We then iterate a) Optionally over different parametric forms

b) For each parametric form over different parameter combinations. The range of parameter

combinations are preferably optimized first according to the spatial distribution of the object points (e.g. limit ellipsoid form's major and minor axis length according to spatial extent of the object points), then compare each model with the object points. The comparison method include for example calculate the distance between model surface and object space points, preferably using their mean, or the mean of their square, or the mean of their absolute values, or a weighted mean of them.

After finding a model matching the object points, the droplet volume is then directly calculated from the model.

Alternative to Ray-tracing

As having been said in _^ Note on Camera Model, the function of ray -tracing might be achieved using mathematical models of the lens system and likely with reduced complexity and increased speed. Therefore, we extend ray -tracing in all this application to include the object space→ image space mapping using the mathematical model (frequently matrix/vector multiplication).

The Multi-Camera (all discussion below applies refer to two or MORE cameras) Approach In ^Locating we described the use of a secondary camera in helping to find the "depth" of droplet tilting in Z axis. There two cameras placed orthogonally were used to capture XY and YZ plane images of the droplet respectively, and the true (Χ,Υ,Ζ) location is found by solving the equations.

Yet another way of finding the true (Χ,Υ,Ζ) location is by:

1. Finding center (centroid) of droplets from the respective images.

2. Reversely trace the rays, and the intersection (close enough) of two rays is the droplet' s true center location.

The idea is also triangulation, and can naturally extend to three of more cameras. Multiple cameras can participate more in addition to finding droplet's location:

Ordinarily, the multiple cameras are synchronized so that their images are taken at the same instant provides different observations of the falling droplet; alternatively, the multiple cameras can be "time-shifted" to increase the frame-rate. For example, if the frame rate of camera_! is 50fps, three such cameras temporally evenly separated achieves 150fps rate. One of its advantage is that because the falling "angular speed" of the droplet in the view of a close-range camera is actually very large, a frame rate of 50fps might not always catch images at the location near the optical axis height (which in general is the best observation area, having smallest distortion, corresponding best to the Fresnel illumination model as illustrated in Fig.C.3-4, etc.), but a lOOfps or 150fps could allow this, then the actual image used to analysis will be chosen to be the one in which the droplet is closest (or relatively closer; certainly not the farthest) to the optical axis. We point out specifically that this is an important use of the multi-camera configuration.

For the ^reverse parametric curve fitting problem: two or more cameras from different viewing location (no requirement on being orthogonal; can be both on YZ plane) can provide different "profiles" of the droplet. The fitting would be that the attempted coefficients try to minimize fitting errors simultaneously on two or more images planes, such a stronger constraint potentially would further increase fitting accuracy.

For the ^forward fitting problem, the fitting is also extended to profiles / object space points/curves observed from multiple cameras at different viewing locations, The fitting would be that the attempted coefficients try to minimize fitting errors simultaneously on the combined profiles/object space points/object space curves, such a stronger constraint potentially would further increase fitting accuracy.

For the image space→object space mapping problem (before §The reverse fitting problem), all methods described above and summarized in Fig.C.6-1 and Fig.C.6-2, of course including the simulation library (virtual/software and real) methods, can be applied. For the simulation library methods, the parametric fitting on object space points, the parametric fitting to image with illumination ray -tracing, and all methods so far discussed, observations from the multiple cameras can be used and cross-checked/verified/validated/compared together to more accurately fixing location, orientation and model parameters (eccentricity of ellipsoid, coefficients of cosine series, spline, NURBS, etc.). There are a variety of ways to treat the results from multiple cameras. One is that we get from first camera volume,, from second camera volume ₂ and the true volume can be determined from these values (such as a weighted sum; or "majority vote" in machine-learning terms), and which values takes higher "weights" is primarily determined by the quality of its image (particularly for "time-shifted" images from which one might chose the better one(s)). We may also directly pick some and discuss others, and derive the true droplet volume from the picked ones.

We also make it clear that all the methods discussed so far for using multiple cameras applies to both 1) multiple view location 2) multiple time instant (see "time-shifted" approach). We make it clear this is the our fundamental definition of "multiple", hence it also includes mechanically changing the relative location/orientation between the drip chamber and the camera (such as motors rotating/moving the camera(s)).

Information within Image

1, B .sr: Difference with edge Mm ami edge dentification

It has been shown in §2. Calculation that y_true · 4% · Tm(\ 0°) = y_trm · 0.705% gives a limit for Z position estimation error for our real lens. In fact we have tried dozens of lens designs to fit within space constraint (device size), F# (f-number), spot size and MTF requirements, and all realistic lens exhibit similar characteristics. Therefore, for low y_trae value (see Fig.C.4-1, near center), the particular accelerometer + trigonometry measurement, and reasonably many of our proposed locating methods, achieve high accuracy in finding droplet's location.

The contents of the present application has so far all centered around optical correction, more particularly, magnification ratio correction, in both microscopic and macroscopic manners and by the fitting method. There is another problem we could not overlook: the edge blur.

Example of blurred spots are shown in Fig.C.4-4, in which we see that there are two types of blur spots: 1) truncated- cone shape for closer objects and 2) sandglass shape for farther objects. Because droplets' dark edges are formed due to Fresnel reflection (see Fig.C.3-4, only "tangent" rays passes without being overwhelmingly deflected), for droplets at the intended "7mm plane" the transition between bight background and dark edge is very sharp, but for other droplet positions the unfocused rays in Fig.C.4.4 shines partially into the originally dark edge, cause them to become brighter and make the transition less sharp.

Since both blur and magnification ratio (pixel-physical distance mapping) change occur simultaneously for droplets off the "7mm" plane, which one is actually responsible for the change s in measured droplet volume? Can the correction of one simultaneously fix problems caused by another?

A quantitative comparison is shown for the quarter 6mm> 6mm area on the XY plane illustrated in Fig.C.4-5(1). We sample grids over this area, recording each grid's 1) absolute RMS blurred spot radius and 2) relative change of image point position's distance from the image point's position when at the "7mm" plane. Z location was taken to be at 5.5 to 8.5mm with 0.5mm increment, and the results are shown in Fig.C.4-5(2).

At all distances, except for when the object is at the central ~lmm area where the image point's distance from the center is very small (this area could be ignored since standard 0.05mL spherical droplet has 2.285mm radius), for farther objects locations the relative image point offset (from the corresponding image point when the object is point is at the "7mm plane") is always several times larger than the absolute blur radius. This highlights the first importance difference in magnitude between edge blur and image point move due the Z offset of the droplet.

The second important difference is that blur always encroaches into the edge peripheral because it unfocused rays centers around the chief ray, whose projection define the "true" edge, and this encroaching in theory pushes the edge inward, though it can be corrected using edge detection algorithms; the actual image point offset, on the other hand, moves outward when droplet moves closer to the lens, and inward in the other direction, which is bidirectional as opposed to the unidirectional inward encroaching of the blurring.

These two important differences, particularly the one on magnitude, having been shown to be generic on all types of optical lens we have designed for the device, and they suggest that any correction of the edge blurring alone cannot concomitantly correct the magnification ratio change. Therefore, edge enhancement/detection which identifies the true droplet edge position in the image is a preferred (not imperative) image processing step, but the magnification ratio/volume correction as discussed extensively above, is always indispensable.

Because blur sizes can be calculated (or measured as discussed below), we further propose a method for blur processing/edge detection:

1. First, refer to Fig.C.4-6 which illustrates the microlens over the sensor array (Omnivision), because of the refracting effect of the micro lens, rays with different incident angles within a cone of considerable solid angle would not have significant difference in their response strength. If this being the fact for any particular lens and camera combination, then it would to some extent simply the work. But the general approach is either

a. Using the (large) lens model, the microlens model and sensor characteristics (response curve relationship with angle and spectrum, Bayer/CYGM/Feveon X3/panchromatic/Fujifilm

EXR/Fujifim -Trans filter, etc.) to calculate, for each possible point source position in the 3D space, corresponding to the possible droplet position due to tilting, drip chamber manufacturing, misalignment in attaching chamber to the device, etc., for different strength of light (not necessarily within visible spectrum), calculate the energy distribution (and response strength, pixel intensity, etc.) over the area centered around the image point location. In one embodiment, the point source light(s) are directional lights so a significant portion its(their) energy is projected into the lens system and eventually hits the sensor therefore makes energy calculation more accurate; in another embodiment, the point sources are allowed to rotate (in addition to move) at each location, therefore its different aiming angles simulate different ray within the ray bundle, and the calculation also incorporate this directional information.

b. Replace the virtual light sources in (a) above, and the microlens/sensor model with real camera mounted behind the lens, with all other characteristics of the above characteristics of the method retained. In one embodiment, the light sources is simulated by a pattern of dots on a surface (can be planar), and surface moves in a 3D spatial region whose trajectory including movement in the Z direction. The dots can be holes acting like "sifts" to allow lights to pass, but they can also be dark spots over bright luminous surface. In a preferred embodiment, the dots, either dark or light passing holes, are etched over a reticle to achieve high precision. The rotational point source light(s) as mentioned in (a), can be implemented by larger rotational non-point source lights, and the sifting property of the dots/holes make it appear like a grid of point sources, this is illustrated in Fig.C.4-7. Of course, the light sources/surface structure (sift, reticle, etc.) are allowed to move, particularly in the Z direction.

The reticle + rotational backlight can also be used as an enhanced version of the Z-shifting calibrating grid in Fig.C.2-4.

In all embodiments, just as in Fig.C.2-4, one can place drip chamber or material with similar optical properly around the light sources (all types mentioned above).

2. The data obtained from (1) in theory is high dimensional data, which include at least{object's spatial location, light strength, light direction, emitting light geometric distribution description (for example, viewing angle at 50°) , 2D area location (centered at the image point) and shape description at the sensor plane, the energy /pixel intensity description over the 2D area} . An optional processing step (can be left to real processing by processor, and this allows calibration after the device has been used for some time and the light intensity degrades/changes) can be applied to reduce the data to a condensed form in the form of a database(s), matrix(ces), formulae(formula), machine-learning/statistical description,. In the extreme the dimensions of spatial distribution might be reduced to depend only on Z change.

3. In field use, after acquiring the image of a falling droplet, we use any of the locating methods described in this application to find the location of droplet (X, Y and Z all affect blur size, as can be seen from Fig.C.4- 5(2), so any or any combinations (for example combining XY into radial distance R; optionally omit Z assuming it is fixed, etc.) of them counts), and retrieve from the above representation characteristics related to the blurred edge (such as information related to the edge depth, edge 2D/radial intensity profile), and uses edge enhancement/detection/identification methods which makes use of such spatial-dependent information, to find the "true" edge which corresponds closest to droplet' s edge. The defining characteristics of the edge- finding algorithm is that does not only rely on the "local" image characteristics itself, nor merely "global" as considering only the image, but including pre-obtained/calculated (can be calculated online) knowledge strongly related to the spatial location (X and/or Y and/or Z) of the droplet/image. This makes it an spatial- aware edge-finding algorithm.

As having been shown above, spatial changes of magnification ratio have stronger influence on volume measurement. So the method here is optionally and preferably followed correction methods in this and the 056090 application.

Because blur size (radius) has been shown in the above discussion to vary with different (Χ,Υ,Ζ), we can reversely infer the (Χ,Υ,Ζ) location of particular points (such as a point on the droplet edge) from this information. In Fig.C.4- 5(3), the right intensity plot is from another image in the same series with the left image but they show good correspondences. Characteristics of the sharp "cliff corresponding to the transition from bright areas to outmost (and other parts, no intention to be restricting) dark edge can be readily analyzed to detect the size of the blur fairly accurately because of the high resolution of the image (5MP OV5640, 1.4μπι pixel size; the fabrication technology would soon allow sub-micron pixel size), and by mapping to Fig.C.4-5(2) (pre or instantly calculated, preferable with tolerance-estimated lens model/transformation/projection matrix), from the point's location in the image plane we can infer its object coordinate.

There is a question that for a particular blur size, say 6μπι, there might be two locations along the point's reversely traced line, one farther than the designing "7mm plane" and one nearer, both having the same blur size. There are a variety of ways to further identify which of the two is. Before explaining this we first introduce Shape Analysis.

Shape Analysis (independent, not merely a sub-section of this section):

Using shape analysis techniques (please refer to [Luciano da Fontoura Costa - Shape Classification And Analysis - Theory And Practice (2ed)]), one could identify the reflective symmetric axis of the droplet's image, its geometric center, major and minor axis and so on (see Fig.C.4-5(4)). One of the uses is that the symmetric axis can be used to rotate the image to up-right orientation. Some operations in the image space, for example, slice-by-slice integration in order to find the volume, works more conveniently with a rotated upright image.

Another use is that this can be used in conjunction with blur analysis to identify point's object location. This is because according to Fig.C.4-5(2), the blur size also varies with the location of the droplet, although not monotonically. For example in Fig.C.4-5(4), along the reversely traced ray from A_b two locations, one farther to the camera and another nearer, could result in the blur size at A_! as found in the image, with which constraint we reduced the search domain from one line to two points. We might further use the blur-size constraints on A₂, B_{1 2}, Ci ₈, and because optically it is extremely unlikely, if not impossible, for the two possible locations on either sides of the "7mm plane" determined from A_{1 ;} to simultaneously yield exactly the same blur size at all the other points, there is a very high probability that the droplet's location can be uniquely determined. This is particularly true when the falling vector is out of the YZ plane so that due to tilting (like the "rotation" effect) points symmetric about the major axis (e.g. B_!→B₂, Ci→C₈) would have different Z coordinate (which is the most important influencer of blur size).

If on the other hand the two locations are indistinguishable from blur size information in the current frame of image, we would analyze subsequent frames until we could be certain of that. The bottom line at the worst case is still that blur size analysis at least allow us to restrict possible droplet locations to two points, and uniquely identify one with high probability. If it turns out insufficient, then the two locations must be used with other analysis techniques to pin down the true location and falling orientation (because once we identify any of the droplet's location, we identify the tilting orientation at least for a period of time, and this location- orientation process is done continuously); if it is sufficient, then the task is completed.

We have therefore shown how the blur size, which initially seeming undesirable, could be used to in finding the true location of the droplet. First we propose a list of image characteristics which could aid in locating droplet:

1. blur size (discussed above)

2. edge characteristics, including gradient, width, intensity, etc. (see Fig.C.3-2 and Fig.C.3-3)

3. magnification (pixel→distance, overall)

4. image pattern: particularly as seen in Fig.C.3-2 and as we have already discussed, the brightness of the

droplet center is due to low Fresnel reflection at small incident angles. At different location, droplet center will allow rays from different part of the illumination to pass through. And because illumination sources (LED, etc.) ' s light intensity in general varies with direction (even for uniform culminated beam, at different distances from the Z axis the amount reaching the image plane are also different), this could be used in determining the droplet's location, particularly in terms of its relative orientation and direction to the lighting sources.

As having been discussed above, A_! and A₂ are major axis endpoints of the droplet. There corresponding object space points 0_Ai and 0_A2 and the breaking point (O^^) near the drip mouth will be collinear. We can trace rays backwardly from A_! and A₂, and because all lens have distortion, the two backwardly traced (or use behavioral mathematical model, or matrix) rays L_! and L₂ are (passes 0_Ai and O_A2) in general are not strictly coplanar. If we connect a line from O^^ to LI which correspond to the physical falling off direction, for realistic droplet location we should expect the extension of this line to also cross (or get very close to) L2. Since LI and L2 are in general non coplanar, L2 will intersect with the plane formed by lines passing O_break and having another point sweeping along LI, and near this intersection will be the droplet's true location. Because we use 5MP (OV5640), and will soon allow lOMP, 20MP and higher camera to capture a small area, the resolution will allow us to do this analysis.

On the other hand, even if the distortion is low and we do not have a very high certainty on the location (a range) given by the above method (L2 crossing plane O-Ll) alone, we could still use the edge characteristics, blur size (a strong parameter), magnification and image pattern (brightness, particularly center, etc.) as additional constraints in determining the droplet location.

Additionally, we could identify falling orientation (and hence location of droplet in every frame) from a sequence of images. For a single image from one camera, in general we know the directional line (by backwardly tracing, or matrix method, etc.) on which the droplet is on. Using the restraining image characteristics discussed above, for each image we could narrow possible droplet location to a small range (such as two positions using blur size analysis). For a sequence of images imagel, image2, image„:

1. the locations must be close proximity

2. the locations and O^^ must be approximately colinear

3. along traj ectory candidates the relative image size (relative magnification between images) must be

consistent with magnification ratio (the gross statistics, can be calculated from individual pixel-distance ratios at spatial grids) change across locations.

4. and at candidate trajectory all the image characteristics discussed above must be continuously consistent. Therefore, from a sequence there is a wealth of information/constraints we could exploit to narrow image location. They are in general done by

1) Take a sequence of images

2) For each image, using backward ray tracing and the image characteristics discussed above to narrow location range

3) For the image sequence, using the constraints (collectively) above to narrow and identify falling direction

We also make it explicit that this locating (image sequence narrowing) method also applies to the case when multiple cameras are used. In addition, the image sequence joint narrowing scheme also applies to other points of the images (minor axis, evenly spaces (along reflective axis direction, or perpendicular to it) rows (particularly ends), columns (particularly ends), etc.)

Error Indicator

It has been shown in 056090 and ttis application that due to tolerances in manufacturing and assembly, different drip chamber types and tilting, there can be deviation in the measured droplet volume from the actual droplet.

It is hereby provided a mechanism, in

1) Visual form (screen, characters, figures, error bar/indicator on screen or as lights/dots of lights, mechanical dial/dashboard, etc.)

2) Audio form (speaker)

3) Signal coupling (electrically, electronically, wirelessly, magnetically, optically, etc.) to a separate device (such as a monitoring center, or error indicator in close proximity) so that regardless of whether the error is corrected or not, the estimated volumetric deviation is conveyed to the user (patient, nurses, etc.). Referring to Fig.C.2-2 for example, the user is hence informed of the error rate either like "the current volumetric drip monitoring/administering rate is x%", or "the cumulated error rate is x%".

Part D, Mechanical Droplet Removal

We refer to the following publications:

[1] [The effect of polymer surface on the wetting and adhesion of liquid systems - J. Adhesion Sci. Technol., Vol. 21, No. 3-4, pp. 227-241 (2007)] [2] [H. Murase, K. Nanishi, H. Kogure, T. Fujibayashi, K. Tamura and H. Haruta, J. Appl. Polym. Sci. 54, 2051 (1994).]

[3] [H. Murase and T. Fujibayashi, Prog. Org. Coatings 31, 97 (1997).]

[4] [E. Pierce - Understanding of sliding and contact angle results in tilted plate experiments, Colloids and Surfaces 2007]

[6] [R. N. Leach, etl., Dropwise condensation: Experiments and simulations of nucleation and growth of water drops in a cooling system]

[6] [Ho-Young Kim, Drop fall-off from the vibrating ceiling (Physics of Fluids 2004)]

From among the above literatures we note:

The sliding angle, adhesion and surface tension are due to interface balance between solid, gas and liquid which are governed by laws in physics, surface chemistry and thermodynamics, hence are predictable.

Hence for the specific type of drip chamber material and IV solution (which might vary in density, viscosity, etc., which can be read from like barcode, marking or user input) we might calculate relevant key property, among which a) Droplet size/volume distribution and spatial density (separation) distribution

b) Resonant frequency (of droplet, of chamber material (the presence of droplet might alter properly, but are all predictable), of part of the drip chamber particularly including (or belong to) the viewing window)

Forced vibration/oscillation is rapid process which for example can break glass cup in seconds with sonic wave. Hence we may rapidly remove (either or both) droplets without interfering with monitoring process. Such a forced vibration may be exerted upon:

1) the drip chamber

2) (either/both) viewing window & droplets, depending on which responses more strongly to the driving force. The force shall be given:

1) Directly mechanically, including parallel shifting, fixed-pivot vibrating, knocking, shown between Fig.D.3-1 to 3;

2) Via medium. The most typical medium is air in which one might apply air pressure (in general, including acoustic pressure, e.g. Fig.D.3-4). Such pressure might be also applied via pressurized nozzles (including fan). Please note that air pressure forcer, acoustic pressure generator, piezoelectric material ,pressurized nozzle might also engage in close distance with the drip chamber/viewing window, in which case the line between "direct" and "indirect" is blurred.

The forcers might intelligently adjust its output upon the following observations: 1) The camera, weight sensor or other may determine the surface level, and coupled with density, the weight, viscosity and other properties of liquid, which affect the two properties a) droplet size/spatial distribution b) resonance response.

2) The forcer might be able to selectively apply influence on specific region, for example, it might consist of an array of pressurized nozzles and activates one to remove the most severe droplet. They may also change its targeting area in ways like rotating RADAR or electronic fan.

Would the vibration affect optical measurement? Refer to the study in § A.2.5 Analysis in which we see that the radius has influence secondary to droplet distance D, which in this case is largely unchanged. According to Fig. A.2.4-5 even for radius change from 6mm to 8mm the measured volume changes only about 3%. when viewing window is being vibrated the magnitude is limited by its bounding areas which can only result in very small optical distortion, and also being momentarily, so the effect on measurement would be very small. When larger region is being vibrated such as the drip chamber is being moved as a whole, more careful control shall be exerted. In the case of Fig.D.3-1 or Fig.D.3- 2 we might prefer vibrating dominantly in direction perpendicular to the optical axis to minimize object distance change.

To further minimize optical distortion we propose we propose using a different curvature at the viewing window. Note that the result in §A.2.5 was reached with Cooke triplet which might change with lens design, so in general we propose that the radius of curvature at one or more points of the viewing window be more than 3% (electable in 1% increment upward) larger or smaller than (either rl, r2 definition in ^Facilitating cooling) of the corresponding cut profile. In most cases we would prefer a "flatter" viewing window with a larger radius of curvature according to the definition.

To intensify vibration two improvements are proposed:

1. As shown in the upper part of Fig.D .3-6, one or more properties which are relevant to the strength of

vibrational response, including but not limited to,

thickness (intuitively straightforward)

density per area (also intuitively clear)

tensile strength/strain

elasticity, elastic stiffness

hardness, either one or more{ scratch, indention, rebound}

ductility

plasticity

toughness

viscoelasticity

, be different from the surrounding area. Relatively we characterize this difference to be greater than 2% and allow electing in 1% increment upwards. Clearly we are trying to make the material "non-homogeneous" in which the viewing window has the strongest response. Materials chemically different shall of course be allowed. 2. From the perspective of its effect. The per-force-per-area-normalized vibrational response strength (the frequency be one calculated to give efficient resonance) of the viewing window be more than 2% (allow electing in 1% increment upwards) greater than the surrounding area. Such a metric is most frequently characterized by response magnitude.

3. The lower part of Fig.D.3-6 shows that we may also intentional create "loose" inhomogeneity at joint between viewing area and surrounding material, such a graphically "thinner", or effectively (according to the host of concrete metrics above or the normalized response metric) "looser" so that the momentum of the viewing window / droplets is effectively detached or isolated from surrounding in much like a trampoline, resulting in stronger response. The same 2% numeric criterion as above paragraph applies.

According to the study of Kim [6], for droplet of 8mm³ volume (1.24mm radius), resonance frequency ranging around 25Hz and 75Hz easily disengages droplets. We shall similarly calculate/experiment/store in firmware for each particular {liquid, chamber/viewing window, droplet size} combination the optimal frequency (or its range).

Fart E. Calibration Library & Dust Removal

The new content in this application is that it addresses issues of lens and camera calibration due to

manufacturing/assembling tolerance. It also discusses how to remove dust accumulated on the lens over time by automatic mechanisms.

This application discusses camera/lens calibration for an IV monitoring device. In particular, we show that a library of calibration backgrounds could be built to achieve accurate droplet volume measurements for drip chambers of different radius, thickness and material index of refraction.

Bs-kf Dess;Hpsios5 oi ihs Dnw issgs

Fig.E.l shows a Triplet (Tessarform) design for droplet volume measurement. Fig.E.2 shows the use of Zemax's PMAG operand to achieve a desired magnification ratio. Fig.E.4 shows the steel ball calibrator and a real or simulated drip chamber used for calibration. Fig.E.5 shows patterns for calibration.

Fig.E.6 shows the use of linear pattern calibrator together with a real or simulated drip chamber for calibration. Fig.E.7 shows the use of circular pattern calibrator together with a real or simulated drip chamber for calibration. Fig.E.8 shows the use of object actual dimensions together with a real or simulated drip chamber for calibration. Fig.E.9 (1) shows a calibration reticle; Fig.E.9(2) illustrate the "calibration background library" concept. Fig.E.l l shows mechanisms for automatic dust removal.

The principle and practice of volumetric measurement of IV drips has been described in great detail in

PCT/IB2013/056090 (the application was over 100 pages). There we have shown an exemplary Cooke Triplet lens system specifically designed for such a system, and have studied the effects of changing variables including

1. D: (optical) axial distance between the center of the droplet and the first surface of the lens system.

2. R: PVC drip chamber's inner radius

3. T: PVC drip chamber's thickness

The experiments are shown in Fig 2.4-l(Fig.A.2.4-l) to Fig2.4-7(Fig.A.2.4-7) in the PCT/IB2013/056090 application, and we saw there that D has a major effect on the measured droplet volume, R has the weaker effect then D, and T the least. The general influencing patterns of the variables, especially T, is complicated, and it best studied using computer simulations as shown in that application.

The ways of computing droplet volume was using the fundamental slice-by-slice integration:

The shape of an improved Cooke Triplet (Tessar form) is shown in Fig.E.1. The last compound lens provide some additional ability in correcting chromatic and other aberrations that is hard to achieve with the basic Triplet form.

It was not difficult to design "perfect" lens for this close-distance photometry application because we only measure droplet volume when it falls to the center of the view so that it is in the paraxial region. In the paraxial region, all optical aberrations can be controlled to be small or even negligible. What is critical to the accuracy of the measurement is therefore the magnification ratio at this region because:

_ 4 ,3

~ 3

So that deviation in magnification error due to manufacturing tolerances will have a cubic effect on the volume, for example 2% increase will cause 6.121% increase in measured volume. Mechanical infusion pumps typically have an accuracy of 5%, so if our device is to compete with them the magnification error needs to be controlled tightly.

During lens manufacturing there are tolerances in

1. Radius

2. Diameter

3. Sag

4. Thickness 5. Air spacing

6. De-center, axis alignment

7. Tilting

8. Index of refraction

9. Abbe Number and they collectively can have a significant effect on magnification. In addition, there are also tolerances in assembling the camera module (focusing, alignment) as well as assembling the camera module into the device housing (will affect the distance D term mentioned above). In general there is a need to calibrate the system after it has been fully assembled.

Fig.E.2 shows an example of intended paraxial magnification ratio. The camera sensor is 5MP OV5640 with image area 3673.6 μπι x 2738.4 μπι. We would like to project a height of 20mm of the drip chamber into the 3673.6 μπι dimension of the sensor, so the target magnification ratio was specified as 0.18368 with lens design software Zemax's PMAG (paraxial magnification) operand. After optimization along with other factors, a 0.1841 (minus sign because the image is inverted) magnification ratio was achieved.

If due to manufacturing, assembling and other tolerances the value PMAG changes to, say, 0.19, then the device needs a means to know this change and compensate for it by (equivalently) multiplying any measured linear dimension with

0.1841/0.19=0.968947»97%.

Calibrators

To better simulate the setting of a droplet behind the PVC chamber (effectively forming a lens, see section 2.2 in PCT/IB2013/056090 "Designs for a very precise IV monitoring system based on computer vision technology"), we could put an actual PVC chamber (see Fig.E.4) in front of the calibrator. The PVC chamber is preferably made of PVC with a low degree of plasticizer so that it is not flexible and has a definite shape (curvature) or even other types of plastic/glass as long as its optical properties (index of refraction, Abbe number, partial dispersion) match or are close. For example, quartz has index of refraction 1.544296 which is very close that of PVC drip chamber material, and they can be manufactured to the thickness (0.5mm) similar to those of drip chamber with high precision.

If we could put an object of known size in front of the camera and examine its image, the actual magnification ratio can then be calculated. For example, this could be an equivalent of a precision ruler with ticks on it, be of either grid, (concentric) circular or linear pattern (see Fig.E.5). These patterns can actually be marked to an extremely high precision using laser engraving or similar technology (possibly lithography on specific materials), and the material to be marked is not limited to metal, but can be any kind of material as long as it has enough rigidity (not to deform) to retain the shape of the marking.

The required "object of known size" can also be an object without special marking. We show two examples:

1) A steel ball (for rolling, bearing, etc .) (see Fig.E.4. The left of Fig.E.4 tries to depict what a real droplet might look like, and the right shows a steel/metallic ball). These balls are not manufactured by grinding, but are instead made by a systematic process involving some physical/chemical treatment, and are produced in large quantities. There precision in terms of diameter tolerance is extremely high (0.5 micron for AFBMA 20 grade). In addition, their spherical shape makes them good substitute calibrator for real droplets.

2) Other shapes other than steel balls, for example a rectangular bar, as shown in Fig.E.8. Precision grinding can make such small parts' dimensions to micron precision and they can also serve as good calibrators.

Still another recommended type of calibration object is reticle. An example design is shown in Fig.E 9(1), in which the most area of the central region of the glass is covered by chrome plating (less than 150 nanometers thick) so that light cannot pass through. The holes on the chrome plating can be made to have diameters of a few micron, and the image of these holes will be formed on the image plane (camera sensor or equivalent) as a dot with smaller (usually) diameter plus the airy disk fringe formed by diffraction, the radius of which could be accurately calculated.

The reverse side of the reticle (not facing the camera) can be sandblasted to make it like a frost glass ("¾¾¾") so that light entering this air-glass interface will be scattered in all direction even after they come out through the holes on the clear (not sandblasted) side, so each hole behaves like a diffusive source. If we do not sandblast the reverse side, particularly when using a collimated beam to illuminate the reticle from the backside, the outgoing rays would largely still travel in parallel directions so that they may not correspond to the ray directions emitted from real objects in the calibrating plane (for example, real objects will emit chief ray passing through the center of the pupil which is not the direction of a collimated beam). If we are not using backside collimated beam for illuminating the reticle, but instead sources like an LED, the outgoing rays would have direction shaped by the illumination pattern of these LEDs, and sandblasting the reverse side cures this effect.

The spacing of the holes in the reticle can be made according to a predetermined pattern. Advanced manufacturing technique can make their positioning accuracy to within ±0.2μπι. The camera would shoot an image of the reticle, storing in processor's firmware as a "calibrated background" in which the accurate positioning of each "dot" is known. When it shoots the image of a real droplet, the internal algorithm will overlay the droplet image with the calibrated background and accurately determine the real geometry of the droplet, including using sub-pixel interpolation, to a very high degree of accuracy.

Calibrated Background Library

As we have shown in our PCT/IB2013/056090 "Designs for a very precise IV monitoring system based on computer vision technology", in chapter 2 "Optical (Lens) Correction", the measured size (affected by magnification ratio and image quality) is affected in an order of decreasing importance by:

1) D: The distance (D) of the droplet's mid plane from the front inner surface of the drip chamber

2) R: PVC drip chamber's inner radius (R)

3) T: PVC drip chamber's thickness

"D" is outside the purview of calibration. For R and T variation, in manufacturing of this IV monitoring and controlling device, we might repeat the above calibration process (using either reticle or other object) with simulated/mimicked drip chamber of different R and T combinations, for example using the same reticle pattern put within PVC or quartz drip chambers of different R and T combination, saving the multitude of calibration grids in the firmware. When the device is working, the R and T canbe measured using methods disclosed in PCT/IB2013/056090 "Designs for a very precise IV monitoring system based on computer vision technology", read (by barcode reader or by camera) from markings such as barcode or equivalent (like two-dimensional code), input by user from keyboard, touchscreen, voice or so on. The device will then trying to find the calibrated ground with the closest R and T setting to the present drip chamber and use that to calculate the droplet volume.

Still another dimension is the drip chamber material's index of refraction. Our study has shown that deviations even in the order of 1/100 from PVC's standard index of refraction have only very limited effect on measured droplet volume. However to accommodate possibilities that drip chambers of different materials might be used (for example TPE plastic, PVC with different type of concentration of plasticizer, glass, etc.), we might add this dimension to the calibration process so that the final product stores in its firmware a cubic array of background grids corresponding to different {R, T, index of refraction} combinations.

An illustration of the "calibration background library" concept is shown in Fig.E.9(2).

On device & Off device calibrators

The calibration can be done in two ways:

First, the device includes such calibrators within itself. Numerous simple mechanisms would be able to move the calibrator into and away from the view of the camera, so we do not bother to discuss it here. Even with the simulated drip chamber (we don't need the full height of the drip chamber, but only a small section of it to simulate the refraction) which could occupy some space within the housing, simple mechanisms could also move it into or away from the view of the camera. So this comprises a self-calibrating optical monitoring device, and the device can periodically during its lifetime do the calibration or after events such as the accelerometer sensed some shock.

Second, the calibrating units (all types mentioned above, but preferably one with a simulated drip chamber like Fig.E.4, because it can be easily mounted into the device) could be provided to hospitals. The device would have some means (automatic recognition, or via user interaction) to know that a calibrating unit is inserted and then uses this unit to do the calibration. This is also very practical because each hospital needs only a few such calibrator units to calibrate all their monitoring devices. This comprises an after-manufacturing-calibratable optical monitoring device.

And of course, the use of said separate calibrating units to calibrate the IV monitoring device also constitute a "process" or "practice" that we would like to claim.

2. A omat c Dos Removal

Although the lens is always put inside the housing which effectively forms a large protective "cover" for it, overtime particles and dust could still accumulate on it and they could cause error to the measurement. Expecting users (nurses) to open the device to periodically clean the front lens is impractically because:

1) This could cause addition alignment or other error to the lens system. 2) This could accidently scratch the lens surface.

Therefore we propose adding automatic lens cleaning system into the device. The basic cleaning method include

1. A vibrator (such as piezo crystal ultrasonic vibration), which vibrates the lens system at a frequency so that the dust can be shaken off without cause adverse effects to the lens system. Similar uses has been found on Olympus SSWF (Super Sonic Wave Filter) for cleaning a "shield" before the sensor because dust can enter the sensor due to repeated replacements of the lens.

2. Using air to blow the dust away from the lens.

3. Using a brush driven by a simple actuator.

As an alternative to cleaning the front lens directly, we could also put a permanent "shield" with weak optical functionality (small thickness, low index of refraction) in front of the first lens, and when cleaning we only need to clean this surface. It has benefit for all the three methods listed above:

1. If the "shield" is only loosely coupled (such as connected by a flexible fringe; but sealed to prevent dust entry) with the remainder of the lens system, the vibrator only needs to vibrate it alone to shake of the dust, without the potential adverse effects to the lens system.

2. The shield's surface is most likely flat, can airflow could blow away dusts easier.

3. Larger and flatter area also makes brushing easier.

What distinguishes the dust-removal system from the existing dust removal system on digital photographic cameras is that all existing dust removal system only removes dust on the sensor (because they are designed to work with different lens, and removing and replacing lens can introduce dust onto the sensor), not the lens. Our proposed lens dust removal system solves the particular problem caused by dust for droplet volume measurement, and it differs from prior arts in both structure (cleaning lens, not sensor) and function.

Claims

Claims

What is claimed:

1. An apparatus for optically monitoring IV flow, consisting of a. A digital camera with a lens mounted on top of it, an illumination system for illuminating the droplets, and backend processor for processing images acquired by the camera

b. Heating apparatus mounted in front of the camera's viewing window of the drip chamber for

preventing/removing condensation and splash droplets from obscuring the view of the camera by heating the viewing window on the drip chamber

2. An apparatus of claim 1, in which the heating element heats the area on the drip chamber corresponding to the region of interest of the viewing camera to above the dew point temperature of the internal humid gas.

3. An apparatus of claim 2, in which the center of the said region of interest area is heated to a temperature between 26 Celsius degree and 42 Celsius degree.

4. An apparatus of claim 3, in which the heating is done by conductive heating from a heat conducting or generating element having direct contact with the drip chamber, the contacting area is above the liquid surface, and that the element has an open area which expose view of the droplets to the viewing camera.

5. An apparatus of claim 2, in which the heating element is a radiation source projecting no more than 1 65mW/mm² (0.03W/(patch area 4.65> 4 mm²), Fig.A.1.3.4-6(1)) net absorbed power to the drip chamber.

6. An apparatus of claim 3, in which the heating element is a resistive heating element generating blackbody radiation with spectrum-dependent emissivity.

7. An apparatus for optically monitoring IV flow, consisting of a. A digital camera with a lens mounted on top of it, an illumination system for illuminating the droplets, and backend processor for processing images acquired by the camera

b. Mechanism for measuring, or allowing the user to input, the radius of the drip chamber (R) , the

thickness of the drip chamber (T), the refractive index of the drip chamber material (n), and mechanism for measuring the drip chamber's central plane's distance (D) to the lens system in the direction of the optical axis.

8. A method for correcting the measured IV droplet volume variation caused by variations in drip chamber's radius (R), thickness (T), material index of refraction (n) and the distance (D) between the central plane of the drip chamber to the lens in the direction of the optical axis of the lens system, consisting of the following steps: a) Compute the measured droplet volume from the sensor image

b) Obtain the full set or a subset of {D, R, T, n} from user input or use the primary lens's optical measurement to calculate R, use a secondary lens, whose axis is perpendicular to the plane determined by the symmetric axis of the drip chamber and the primary lens' optical axis, to measure D, the ratio between an radiation beam receiver and emitter to determine T using Beer-Lambert's law

c) compare {D, R, T, n} with the original {D₀, R₀, T₀, n₀} for which the lens system was designed, and using a correction factor or matrix to correct the measured volume in step a) to the true volume

9. An apparatus of claim 1, in which the measured droplet volume is corrected for the set of tuples {e, h, θ, τ} containing: eccentricity e of the droplet profile; vertical distance h of the droplet from the optical axis; rotational angle Θ with respect to the plane perpendicular to the vertical axis; tilting angle with respect to the vertical axis.

10. An apparatus of claim 9, in which Laplace-Young equation is used to fit the droplet image profile and the volume is calculated from the fitting curve.

12. An apparatus of claim 9, in which the tilting angle is measured with either a gyroscope of accelerometer.

13. The combination of a video IV monitor with anti-splash drip chamber, of which the IV monitor is equipped with heating capability to heat the viewing window above the dew point of the internal gas, and of which the drip chamber is made such that between dripping rate 20 drips/min and 120 drips/min the design of the drip chamber guarantee that there be no splash droplet on the inner surface.

14. A combination of claim 13, in which the drip chamber is made of material consisting of, or coated with lubricating material including silicone.

15. A combination of claim 13, in which the drip chamber is made of material consisting of, or coated with low coefficient of friction material including PTFE (Polytetrafluoroethylene).

16. An apparatus of claim 1, which includes a vibrational mechanism to force resonance of either the drip chamber or viewing window or droplets for removing droplets on viewing window.

17. An apparatus of claim 16, in which the vibrator is a piezoelectric ceramic transducer generating tunable acoustic pressure.

18. An apparatus of claim 16, in which in front of the viewing window there are pressurized nozzles for generating pulsing air pressure to hit the viewing window.

19. An apparatus of claim 1, in which the lens system includes piezoelectric vibration mechanism such that upon detecting contamination of the lens surface by dust or droplet, the vibrational mechanism activates the shake the lens at resonating frequency to remove the dust and contamination.