WO2022125312A1 - Identification and tracking of infection in humans - Google Patents

Abstract

Apparatus and methodology for measurement, identification, knowledge inference and notification alerting of infections within individuals and populations. Embodiments according to the present disclosure include a system that enables real-time measurement, data delivery, management, processing, analysis, notification and reporting, thereby providing accurate early detection of fevers caused by infections. Population and individual tracking systems are developed using temporal sensor data. Inference of infections are made through customized methodologies for individuals and populations. Accuracy and usefulness of the disclosure is improved through training a system based on populations of end-users. Further, accuracy of the disclosure may be improved by including most recently collected data. Data input may incorporate measurements of heartbeat data or at least one body temperature. System outputs are based upon deviations outside of pre-determined normal acceptable limits for each user. Predicted outputs may include temperature or physiological normality, or direct predictions of fever, or infection.

Description

IDENTIFICATION AND TRACKING OF INFECTION IN HUMANS

FIELD OF DISCLOSURE

The present disclosure is directed generally to detecting, measuring, or recording infections in humans for diagnostic purposes.

BACKGROUND INFORMATION

It has long been understood that pathogenic infection frequently results in a pathophysiological fever response. Fever, while a commonly used term to describe many diseases and physiological conditions, lacks a formally accepted definition. For the purposes of this disclosure, "fever" refers to abnormal body temperatures. In this context, fever is often described as a complex physiologic response to disease, mediated by pyrogenic cytokines and characterized by a rise in deep body temperature, generation of acute phase reactants, and activation of immune systems. Within medical and research environments, deep body temperature may accurately be determined by measuring rectal, oral, aural or axilla temperatures. These measurements may then be compared to standards which set limits for normal or abnormal body temperatures.

Classifying individuals as febrile using genericized absolute temperature limits, requires supporting evidence from additional medical testing. Within populations, normal deep body temperatures vary widely due to race, age, fitness, health, activity, eating and drinking, environment, clothing, sex, menstrual, seasonal, diurnal, and circadian cycles. Characterizations of normal body temperature within specific individuals requires an understanding of each of these attributes and physical phenomena.

SUMMARY OF THE DISCLOSURE

Embodiments of the present disclosure provide the capability of quickly and accurately tracking, identifying, and providing early warning predictions of individual physiological states related to pathogenic infections.

This subject matter relates generally to an infection prediction tracking system. More specifically, the invention describes apparatus and methodologies that use physiological measurements, real-time data transmission, storage, and manipulation, combined with intelligent decision-making algorithms and communication systems. Deployments of this disclosure tracks, predicts, identifies, and allows timely actions based on automated decision-making algorithms for individuals expressing fevers due to infection. Fevers may be identified through abnormal individual body physiology and body temperatures. Actions taken to isolate those individuals who have been identified as abnormal or infected, improves the safety of public and private spaces and can prevent epidemic outbreaks of communicable diseases.

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing aspects and many of the attendant advantages of the claimed subject matter will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

FIGURE 1 shows an example of deep body temperature measured on an individual over a single day;

FIGURE 2 shows an example of deep body temperature measured on a female individual over 30 days;

FIGURE 3 shows an example distribution of oral temperature measurements of many individuals, with an overlay of simulated data from febrile individuals;

FIGURE 4 shows a diagrammatically one example implementation of a workplace infection tracking and identification apparatus;

FIGURE 5 shows a graphical representation of an example of a temperature and physiological data monitoring device for different body regions;

FIGURE 6 is a block flow-diagram of a representative system development for an individual end-user infection tracking and identification apparatus;

FIGURE 7 shows a block diagram representation of physiological, personal, physical, and environmental factors that may be used for development of an infection prediction apparatus;

FIGURE 8 is a block flow-diagram of a representative system development and training methodology for a population infection prediction apparatus; FIGURE 9 is a block flow-diagram representation of a process that may be implemented by embodiments of the disclosure, such as the embodiment shown in FIGURE 4, in which data are collected and analyzed, and decision/actions are rendered based upon results;

FIGURE 10A shows an example of a predictive tracking system development using data from three end-users, measured for 45 days;

FIGURES 10B and 10C show probability of infections from a representative regression-based tracking and identification system for two end-users in FIGURE 10A;

FIGURES 10D and 10E show infection predictions from a representative classification tracking and identification system for two end-users in FIGURE 10A.

DETAILED DESCRIPTION

The detailed description set forth below in connection with the appended drawings, where like numerals reference like elements, is intended as a description of various embodiments of the disclosed subject matter and is not intended to represent the only embodiments. Each embodiment described in this disclosure is provided merely as an example or illustration and should not be construed as preferred or advantageous over other embodiments. The illustrative examples provided herein are not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed. Similarly, any steps described herein may be interchangeable with other steps, or combinations of steps, in order to achieve the same or substantially similar result. Moreover, some of the method steps can be carried serially or in parallel, or in any order unless specifically expressed or understood in the context of other method steps.

Different methods may be used when predicting infection status. One method may measure individual body temperatures and compare these single point measurements to a set of standard values. An example of a fever definition published by the United States Center for Disease Control (US CDC), under multiple codes, identifies a patient as febrile (having a fever), if he or she has a measured temperature of at least 100.4°F (38°C), or feels warm to the touch, or gives a history of feeling feverish. While convenient to describe fevers in this manner, body temperatures, without supporting physical and medical data are meaningless in characterizing febrile responses. Robustly defining thermal febrile responses may involve definitions for body temperature measurement sites, times of day, times after eating or drinking, specified characterizations of menstrual cycles (female), standard sets of environmental conditions and activity rates, and additionally, knowledge of individuals' age, health, acclimatization to heat and exercise, and quantifying differences due to seasonal variations, among other possible criteria.

As a simplification, daily human temperature cycles are often represented using sine waves, the reality however, is that temperature cycles of individuals are significantly more complicated. Demonstrating the complexity of determining normal body temperature; FIGURE 1 shows an example of a single normal (afebrile) individual's deep body (stomach) temperature captured over a 24-hour period, sampled at 30-minute intervals. The 'x' axis displays measurement time, while the 'y' axis displays body temperature. Measured temperature, represented by black circles, varies based upon time of day, environmental conditions, clothing, activity, and the like. While the subject is sleeping, to the left of 110, body temperature plateaus at a daily low value. When the subject rises and performs typical morning activities e.g. washing and eating in the region between 110-120, then body temperature starts to rise. As the daily routine varies, body temperature changes. In this example, the subject performs office work at a desk for most of the working day in the region between 120-130. After finishing work 130, the subject spends a period exercising, in which body temperature rises to a maximum of ~38.5°C 140. In this non-limiting example, body temperature rise caused by exercise is ~1°C. In this example, the peak temperature of ~38.5°C 140, continues to affect body temperature for hours after cessation of exercise activities. As the subject performs typical evening activities e.g. eating, socializing, resting, body temperature falls. After retiring to bed 150, subject body temperature falls back to a low plateau.

CDC febrile guidance temperature of 38°C 160 is marked on FIGURE 1 for reference. No two days for any individual are exactly the same, therefore, cyclical natures of circadian variations also change between days. Circadian variation tracks an individual's sleep wake cycle, which when synchronized with the day -night phase, is termed a diurnal cycle.

Normal body temperature is further complicated in females by menstrual cycle, where mean daily body temperature differences between mid-luteal and mid-follicular cycle phases may vary from 0.3 to 0.7°C. FIGURE 2 shows an example of a female individual measured at 30-minute intervals over a period of 30 days. In this example, deep body temperature was measured in the stomach using ingestible temperature pills. The 'x' axis displays measurement date and time, while the 'y ' axis displays deep body temperature in Celsius. As previously discussed, body temperature, marked with unfilled circles, FIGURE 2, varies based upon time of day, environmental conditions, clothing, eating and drinking, state of health, activity, and the like. Peak temperatures, rising above 38°C may be experienced during periods of exercise. As illustrated in FIGURE 2, body temperature is also influenced by menstrual cycle. Average daily temperature is marked as a solid black line using a 48-period moving average. In this non limiting example, average daily temperature varies by up to 0.5°C, while the minimum (36.2°C) to maximum (38.7°C) temperature range is 2.5°C. Again, the CDC febrile guidance temperature of 38°C 160 is marked for reference.

Female monthly cycle temperature variation may be further complicated by contraceptive methods, pregnancy, and menopause. In a non-limiting example consisting of menopausal females, temperature variation is observed during "hot-flashes", caused by hormonal variations, which last between 15 and 120 minutes and may affect body temperature elevations of up to 1.5°C several times daily.

Additional complications arise when distinct medical and commercial organizations measure temperatures in different body regions, using different procedures, devices and have their own standard threshold values for normal body temperature. Normal human body temperature may be described by the lowest plateau while resting, referred to as Basal Resting Temperature (BRT). As described above, body temperature is at its lowest during the low plateau, close to the middle of a normal sleeping period. Therefore, BRT is rarely measured in practice, but may be approximated by oral temperature sampled close to waking. BRT can be a particularly useful measurement basis for females attempting to become pregnant, as measuring oral temperature just after waking is a simple and repeatable method for individuals with no medical training.

FIGURE 3 displays cohort data from the Stanford Translational Research Integrated Database (STRIDE), collected between 2007 to 2017. This data compiles over 570,000 oral temperature measurements collected from patient encounters of over 150,000 individuals during visits to Stanford Health Care. -Oral temperatures were obtained by trained physicians, under controlled clinical conditions with digital thermometers, calibrated annually. Any observations having a diagnosis of fever at the time of examination were removed from the dataset. Statistics calculated from this dataset include an overall mean body temperature of ~36.7°C, the mean for males and females respectively is ~36.6°C and ~36.8°C (not shown), both have distributions of ~±1°C. Additionally, this data confirms correlations between oral temperatures and the time of day that the measurement was obtained. A summary of the data is provided in FIGURE 3, where oral temperature measurements are first grouped into 0.1°F (0.0556°C) bins, and the number of measurements in each bin are counted to determine the bin frequency. After counts are totaled, they are divided by the bin with the highest count i.e. the mode. This data is summarized in FIGURE 3 as a normalized histogram distribution of oral temperature measurements. The x-axis shows midpoint temperatures of the 0.0556°C (0.1°F) bins and the y-axis displays normalized frequencies of each bin. Black bars in FIGURE 3 illustrate normalized oral temperature measurements. The normalized data bin with the highest count, mode, has its midpoint at 36.6778°C (98°F). CDC febrile guidance temperature of 38°C 190, is marked by the dashed vertical line for reference.

Infrared (IR) thermometers and thermal imagers may be employed in public and workplace spaces to identify individuals with elevated body temperatures in crowd situations. Infrared transducers detect emitted heat radiation, typically between wavelengths of l-14pm. These wavelengths have the advantage of being able to see temperatures. Thermographic cameras or thermal imagers are technologies employing arrays of infrared sensors that measure many emitted points of heat radiation over a field of view (FOV). Visualization software combines measurements from arrays to form an artificially colored image. Images from thermographic cameras use colors mapped to individual pixels to represent different temperatures, that is, different quantities of emitted radiation. Thermal imagers are particularly useful as they measure temperature at a distance i.e. non-contact and can map many temperatures in a single frame. As discussed above, normal body temperature varies throughout the day (and month), therefore, temperatures measured for the same individual early in the morning likely differ from those obtained in late afternoon. Additionally, environmental climates may significantly alter offsets between deep body and skin temperatures.

With supplementary software from artificial intelligence (Al) models, thermal imagers can automatically identify and isolate facial regions and provide average face, or partial face temperatures. Combined with software that adds cutoff temperature thresholds, these cameras may be used to identify individuals with facial skin temperatures above set values. Several limitations exist with infrared sensor arrays that must be accounted for in the accuracy of any predictions. Manufacturing methods used to produce these technologies are expensive and keep resolutions of thermal imagers far below their visible camera counterparts, often by 1-2 orders of magnitude. Observed resolutions from thermal imagers are highly dependent on objective distances, lens materials and quality, surfaces being measured e.g. color, dryness, texture, angle of incidence and the like, and ambient measurement environments. Additionally, offsets between skin and deep body temperatures do not remain constant and may vary greatly depending on environmental conditions.

Thermal imagers may be placed in high traffic access points of public spaces to provide alerts to trained operators when an individual with a specified body region exceeds a pre-defined constant value. This value may attempt to represent febrile individuals with deep body temperatures of >38°C. Thermal Imagers measuring skin temperatures can apply offsets to compensate for reductions caused by skin, fat, and bone, and thereby infer deep body temperatures. Due to the nature of these devices, selection of manufacturer and environment in which they are used, actual temperature cutoffs can be set at many different values, as can offsets for converting skin to deep body temperatures.

By making a few assumptions, it is possible to quantify the accuracies that these devices, or any other single point measurement of skin temperature, can provide. Clinical literature often defines febrile (in the thermal sense) as having an elevated body temperature that persists for an extended period. Febrile temperature elevations normally range between 0.5-2.0°C, though can exceed 3.0°C in exceptional cases. A typical value used for fevers produced by common communicable diseases is an elevated temperature 1°C above normal. Fevers can be acute or chronic, but for most flu, SARS, or Corona type viruses, they typically last between 1-7 days. For this example, an average fever length of 4 days is selected. If an assumption is made that during a disease outbreak, 10% of the population becomes infected during a specified month, with a mean fever length of 4 days, then using an idealized construct, 1.333% of the population is infected at any single measurement time.

If 1.333% of the population summarized in FIGURE 3 is given a fever, raising their temperature by 1°C, then a composite population consisting of both febrile and afebrile individuals can be simulated. Results of this composite population are shown by the white bars on FIGURE 3. Comparing the black to white bars, small changes are observed in the overall data distribution, shifting mean body temperature upwards. TABLE 1

Statistical Results of 10,000 Single Point Temperature

Screening Test Groups on Composite Data from FIGURE 3

Temperature True False False True

Cutoff (°C) Negative Negative Positive Positive

TN FN FP TP

36.5 442 0 9425 133

37 3198 1 6669 132

37.5 8535 5 1332 128

38 9686 44 181 89

38.5 9804 116 63 17

39 9850 131 17 2

Table 1 displays statistical results of applying temperature cutoff thresholds to FIGURE 3 composite data, where results have been scaled to represent 10,000 individual screening tests for each cutoff. As discussed above, for this non-limiting example, 1.333% are febrile, resulting in 133 and 9867 febrile and afebrile individuals respectively during each testing cutoff. In this example, an offset of 0.5°C has been added to composite data to convert oral to deep body temperatures. This offset value has been included for demonstration purposes only. If cutoff thresholds are applied to everyone's measured temperature, then they either fall below that value, i.e. afebrile, or above the threshold, i.e. febrile. Column 1 of Table 1 displays six different upper threshold temperature cutoff values, where deep body temperature cutoffs have been evaluated with a 0.5°C graduation, between 36.5-39°C. These results are split into four groups. Columns 2 and 5 display the correct test results for the composite population, where afebrile individuals are identified as afebrile (True Negative, TN) and febrile individuals are identified as febrile (True Positive, TP). Columns 3 and 4 display incorrect test results for the composite population, where febrile individuals are classified as afebrile (False Negative, FN) and afebrile individuals are classified as febrile (False Positive, FP). Each row of Table 1 sums to a total of 10,000 screening tests for all four groups.

While evaluating this type of epidemiological data, wrongly identifying afebrile individuals (False Positive Rate) are often considered less important than missing febrile individuals (False Negative Rate). However, additional complications arise during analysis of heavily biased populations, in this scenario 98.667% of the population should be identified as negative. Therefore, absolute numbers may be more important than portion of significance, resulting in type I errors (a) having different significance than type II errors (P). Reconciling the issues described above in rows of Table 1, interesting changes occur between 37.5 - 38°C (oral+0.5°C) temperatures. Between these two rows, the ratio of true positives to false negatives, changes from -25:1 to -2:1, and the false positive to true positive ratio changes from -10:1 to -2:1.

TABLE 2

Accuracy Calculations of Single Point Temperature Tests on Composite Data from FIGURE 3

Temperature Sensitivity Specificity Precision

Cutoff (°C) TPR TNR PPV

37.5 96.2% 86.5% 8.8%

37.6 92.5% 91.8% 13.1%

37.7 88.0% 94.9% 18.9%

37.8 83.5% 96.5% 24.1%

37.9 75.9% 97.6% 29.8%

38 66.9% 98.2% 33.0%

Table 2 expands Table 1 rows between 37.5 - 38°C to use 0.1°C graduations between cutoff thresholds. Data displayed in Table 2 uses groups of individual screening tests for each cutoff, but further calculates statistics based on normalized group frequencies. Column 1 displays upper temperature cutoff thresholds in Celsius. Column 2 displays test Sensitivity or True Positive Rate (TPR) calculated by dividing the number of correctly identified febrile individuals by the total febrile population, i.e. TP/(TP + FN). Column 3 displays test Specificity or True Negative Rate (TNR) calculated by dividing the number of correctly identified afebrile individuals by the total afebrile population, i.e. TN/(TN + FP). Column 4 represents Precision or Positive Predictive Value (PPV), which calculates the number of correctly identified febrile individuals as a portion of the total number of positive test results i.e. TP/(TP + FP).

As illustrated in Table 2, the highest accuracy, as measured by precision is 33%, observed in the 38°C row. At this temperature cutoff ~ 1/3 of the febrile population are misclassified as afebrile, and 7 out of 10 positive tests are false positives. Further, this is an idealized best-case estimate for any single point skin measurement test, as these results are calculated on high quality data captured by trained professionals under controlled clinical conditions, they include no variation associated with the measurement itself.

Under real use conditions, uncontrolled climatic environments and physical activities, temperature accuracies of thermal imagers verses calibrated digital thermometers, and offset differences of skin and deep body temperatures both within, and between individuals should be accounted for. It is estimated that when operated in real- use public environments, the highest expected precision of using any single point skin measurement test for fever identification, measured under uncontrolled conditions, is not more than 10%.

One method to accurately gather body temperature data is to carefully measure and record one or more of the medically recognized standard sites over an extended period. This data may then be used to identify a significant deviation, for example >1°C, within that individual, specifying a set time, in a defined environment and under defined activity levels.

As demonstrated above, even with best case assumptions, a technical solution capable of quickly and easily identifying febrile individuals before they become a public hazard does not currently exist. This disclosure describes a methodology and apparatus using physiological measurements — which may be based primarily upon skin temperatures, heart beat data, and/or blood oxygen levels — to track, identify, predict, and inform infection-based decisions and actions. The apparatus enables delayed or real-time physiological measurement, transmission, storage, manipulation, decision-making algorithms, communication, analysis and alerting systems. Predictions and alerts of abnormal body temperature or physiology provide early warnings of individual pathogenic infection, thereby helping to prevent epidemic outbreaks of communicable diseases.

Another methodology for characterizing body temperature may involve measuring heart beat data or one or more skin temperatures over extended periods. Using this data and applying a full knowledge of each individual's medical history, activity, fitness and the like, accurate analogs of deep body temperature may be developed by analyzing historic datasets and customizing them by individual. Predictions of deviations from normal temperatures may also be made from multiple body sites including the hands, wrists, arms, torso, head, neck and the like, using simple devices such as watches, rings, earpieces, head/neck-bands or eyeglasses that measure physiological parameters.

FIGURE 4 provides an overview of one preferred embodiment of the disclosure deployed in a manufacturing type facility 400. This embodiment illustrates a tracking system for a team of four end users 410, each individually labeled 001, 002, 003 and 004. In this embodiment, the users 410 self-select one or more of 3 types of untethered, self- powered measurement devices. In various embodiments, these measurement devices may be implemented as heart sensors that measure heart rate, or heart rhythm, or both; as temperature sensors that measure a subject's body temperature; or as pulse oximeters that measure a subject's blood oxygen level. Alternatively, the measurement devices may incorporate two or more different types of physiological sensors.

In one certain embodiment, three physiological measurement hardware devices offered may include a ring 420, a watch 421, or an earpiece 425. Each of these devices may have the capability to communicate via Bluetooth or 802. lx wireless networking protocols, among others. As users perform their normal daily duties the devices measure and transmit secured data in real-time. Receiving hubs 430 have the capability to receive data transmitted, for example, via secured Bluetooth or by IEEE 802.16 wireless networking protocols. Hubs 430 pass received measurement data to a local data storage server 440, where physiological measurements are stored in non-volatile memory, such as within a processed SQL database, and then transformed and enhanced according to a protocol 450. Enhanced measurement records for each end-user are evaluated by a processor within the local server 440 using compiled predictive neural network models 460, customized for each individual user and previously activated on the local server 440. Output predictions are passed back to the server 440 via path 465 and may be displayed on a software platform portal 495.

Referring to this embodiment, the software platform portal 495 may be available through any combination of local and remote computers, tablets, smartphones, and other devices. A decision is made on predictive outputs according to a pre-determined protocol for the group of four users. In this non-limiting example, if a user prediction falls below the decision threshold 470, that user(s) is classified as normal 475 and no other action is taken. If the user predictive output falls above a predefined decision threshold 470 the corresponding user(s) is classified as infected (shown as a crosshatched region 480 in FIGURE 4). In one example, User 001's data 496 crosses above the threshold 470. Once a user(s) has been identified as infected, the tracking system sends notification alerts 490 to various recipients, such as directly to the affected user(s) 491, to a human resources (HR) safety representative(s) 492, or others. The tracking system may also post a flag directly on the portal 493.

The human body presents many opportunities to measure skin temperatures and heart beat data. In one embodiment, the chosen site(s) are easily accessible in typical employment environments, do not restrict daily working activities and provide good representations of deep body temperature. FIGURE 5 is a graphical representation of preferential sites that may be chosen to meet skin temperature measurement criteria 500. Recommended measurement sites are highlighted by enclosed hatched areas. Deployments selecting body regions that can experience subcutaneous (below-skin) vasoconstriction, must take extra precautions. Body regions may vasoconstrict under typical thermal environments when a body becomes cold, or due to medical conditions, for example peripheral vascular disease (PVD). Cutaneous vasoconstriction, reduction in skin blood flow, occurs over much of the body and does not impede a functioning system. Regions in which deep body vasoconstriction may occur, reducing blood flow to deeper tissues are typically located in the body periphery. These regions include lower legs and feet denoted by areas below line 510 and lower arms, wrists and hands denoted by areas left of line 520 and right of line 525. Measuring temperature at any of these sites may involve special consideration, as vasoconstriction may cause situations in which deep body temperature is increasing, while the temperature of these regions is simultaneously decreasing.

Therefore, preferential sites for single skin temperature measurements, ideally include those regions that do not reduce blood flow to deeper tissues in response to thermal stimuli. In one embodiment, posterior ear skin temperature is measured over or near the posterior auricular branch of the carotid artery 536. Several sites may be available for measurement in and around the ear, an expanded ear is shown in exploded view 530. One such site is tympanic membrane temperature (not shown), though the ear facilitates several other easier to measure opportunities that closely track deep body temperature, these include measurement of outer ear canal or conch temperature, through an earpiece or earbud 535, like those used for listening to music. Additionally, measurement of skin behind the ear, through an outer ear clip 536 may provide temperatures that closely track the deep body due to its position relative to main vasculature i.e. the posterior auricular.

Other sites that track deep body temperature may include forehead 540, which may be measured through a headband or cap, and neck close to the carotid artery, this may be measured through a neckband 550. Further, Inner canthus of the eyes (not shown) provide sites that closely map deep body temperatures and may be measured with a non-contact type sensor attached to standard or protective eyewear. Mandible temperature (not shown) may especially track deep body when measured over, or and particularly near facial or posterior auricular (behind the ear) branches.

Moving down the body, other preferential sites include the chest in the pectoralis major region 560, which may be facilitated through an adhesive patch or clipped under a brassiere strap. Bicep or inner upper arm region 570, measured through an arm-band (or upper inner thighs) are sites that may also closely track body temperature, as can the abdomen 580, where a sensor can be attached inside of the pants or skirt waist-band. Alternative sites that may be better choices for employment environments, but also provide temperatures with higher deviations as discussed above include, wrist, ankle, fingers, and forearms (not shown).

A benefit of measuring vasoconstricting sites in combination with non- vasoconstricting regions is that during febrile responses, the thermal effector zone is reset to higher temperatures. During this response, extremity blood capillaries often vasoconstrict, occluding blood-flow and creating a larger temperature differential between these regions than is observed in afebrile individuals. This larger difference between regions may provide strong evidence of febrile responses in characterized individuals.

Choice and position of skin temperature measurement site(s) can be tailored to specific working environments and functions, for example end-users working on manufacturing lines may achieve adequate measurements whilst wearing wrist devices or rings. While some body sites may provide convenience for device attachment, these should be carefully considered as environmental factors such as clothing or ambient temperature may cause significant variation in measurements.

Another commonly used indicator of pathogen infections, are alterations to resting heart rate and beat rhythm, therefore, similar to body temperature, predictions made using heart beat data may provide early identification of infection. Heart rate and heart beat rhythm can be measured by embodiments of this disclosure. For example, devices equipped with photoelectric pulse wave technology may be used. Measurement from many body sites provide adequate representations of heart beat data, including most of the regions previously discussed for skin temperature and those shown in FIGURE 5. In one embodiment, heart beat measurement site is the wrist, such as through an equipped smartwatch.

Other devices with capabilities to measure multiple physiological and physical phenomena including skin temperature and heart beat rhythm may also be used. While embodiments of this disclosure may be measurement hardware agnostic, a person of ordinary skill in the art can select their preferred vendors with physiological measurement hardware that meets system design requirements.

Physiological measurement hardware that may be used in embodiments of this disclosure may include one or more of the functionalities listed below.

1. Sensor circuits capable of measuring physiological, physical, and environmental input variables. Sensor circuit design and construction may be of analog, digital or smart circuit/module types. Sensor circuit specifications should include traceability for accuracy, calibration, and drift.

2. If not included within the sensor circuit itself, a separate signal amplification, signal conditioning and communication system may be used to interface with the input/output (I/O) interface of the device processor.

3. A power source capable of powering the device for 2 hours, for example, or longer in a manner that does not tether the individual to a physical location is preferred. Devices capable of providing power for 40-hour plus work weeks without recharging are preferred, although this is not a requirement. 4. Physiological measurement devices should contain memory capable of storing data until such time as it is successfully transmitted to additional platforms. The device may use volatile and other types of memory e.g. a Secure Digital (SD) memory card, although this is not a requirement.

5. The device has a processor with an accurate clock, capable of being configured to record physiological data from sensors, and to send data stored in device internal memory at regular intervals. Configurable intervals for recording and sending data may be any interval between 0.001 to 90000 seconds, with typical values falling between 60-1800 seconds. These processors are optionally capable of making SMART decisions to improve data quality according to a predefined protocol, although this is not a requirement.

6. Eligible devices may send data wirelessly without being tethered. Non-limiting examples include Bluetooth, IEEE 802. lx protocols, Near Field Communication (NFC), cellular protocols, Internet of Things (loT) protocols, satellite protocols, other wireless network protocols, and the like. Ideally, devices include a mesh network of several communication methods and use encrypted communication protocols, although this is not a requirement.

Optionally, physiological measurement hardware is tailorable to many different body types, shapes, and sizes, easily recharged and provide methods for simple cleaning and sterilization.

In one embodiment, data is transmitted to personal communication devices such as a smart phone or tablet and then transmitted to a cloud-based server. Alternatively, in another embodiment, data is collected and then sent to a local or cloud-based server through a connected loT type device, or the like.

Similarly, embodiments of this disclosure are agnostic to communication hub hardware, however, when communication hubs are used, these should meet the basic functionalities listed below.

1. Communication hubs capable of being configured by those of ordinary skill in the art, to receive transmissions of at least one protocol from selected physiological measurement devices. Ideally, communication hubs have capabilities to receive and process more than one type of wireless transmission, although this is not a requirement.

2. Hubs can communicate in real-time (or near real-time) with data servers or workstations, through wired or wireless protocols. Ideally, incoming and outgoing communication protocols are encrypted, although this is not a requirement.

Optionally, communication hubs have power backups and redundant systems to improve system up-time, and also have internal memory capable of storing any nontransmitted data until such time as the data is communicated to the unprocessed data server.

In one embodiment, data is sent from physiological measurement devices, through a cellular network and is communicated to an unprocessed data server through a secured internet connection.

Unprocessed data servers may take many forms, non-limiting examples include data-center rack mounted server(s), local rack mounted server(s), tower or blade type server(s), owned or rented cloud-based server(s), professional workstation(s), virtual server(s) and the like. Capabilities of an unprocessed data server are dependent on the type and size of deployment that server supports. Unprocessed data servers have some common functionalities listed below.

1. Server(s) actual or virtualized; capable of hosting and running at least one type of hierarchical, relational, or object-oriented database.

2. Non-volatile storage capacity to store data from a deployment(s) that the server supports.

3. Processing power and Random- Access-Memory (RAM) capable of running a database(s) and processing data from any deployment(s) that the server supports in a timely manner. Ideally, data processing is in real-time or near real-time, although this is not a requirement.

4. Communications protocols and/or hardware, actual or virtualized, which receive data from physiological measurement devices, either directly, through hubs, or from other relay equipment, for any deployment(s) that the server supports.

5. Communication protocols and/or hardware, actual or virtualized, to transmit data in real-time (or near real-time) to other servers or devices supporting deployments. Functionalities for hardware that executes data optimization, transformations and data compression protocols are similar to those for unprocessed data servers. These operations may be completed by the unprocessed data server(s) or by different computers. Optimizations, transformation, and compression should be completed in a timely manner and transmitted to equipment executing customized end-user decision-making algorithms.

Functionalities for computers that execute customized decision-making algorithms and output predictions are also similar to those described above, in that this computing equipment is capable of executing operations in a timely manner and communicating predictive outputs to storage and notification servers. Optionally, processing raw data, data optimization and execution of decision making-algorithms may be run on the same server(s).

Server hardware functionalities for systems hosting user interaction portals and executing notification alerts are dependent on the deployment(s) being served. Nonlimiting examples of portal and notification servers include, data-center rack mounted server(s), local rack mounted server(s), tower or blade type server(s), owned or rented cloud-based server(s), Voice over Internet Protocol (VoIP) server, a subscription based Graphical User Interface (GUI) and/or messaging service and the like. Optionally, this capability can be integrated into the server(s) described above. Basic requirements for all Portal and notification servers include having at least one method of providing notification alerts to customers and/or end-user(s) of end-user and or system status.

In one embodiment, a minimal installation may forego the active portal for viewing current or historical data. In such an embodiment, the system may send regular emails to the customer of the equipment and user(s) status.

Ideally, evaluation of decision-making algorithms and all processing of data prior to this point are made in real time (or near real-time) after receiving physiological measurements, though this is not a requirement.

Server functionalities for training end-user and population models may also depend on the size and type of deployment being served. Ideally, these servers contain hardware specialized for big data machine learning e.g. multi-core, multi-threaded systems with many Graphical Processing Units (GPU) or Tensor Processing Units (TPU) cores, having high RAM capacities capable of loading large datasets into active memory and supporting highly parallelized software to enable fast processing of large datasets, though this is not a requirement. In one embodiment, a server is cloud based and accessed through a customer's internet connection. In such an embodiment, a local server may be avoided, and data and decision-making algorithms may be stored and processed through the cloud server. In another embodiment, data may be sent to a specialized server for processing Al data sets. In such an embodiment, local data storage and servers may be avoided. The user's physiological data storage, optimization, training, tracking, predictions of infection, user notifications, customer notifications and portal operation may be processed through the same server or through some combination of servers. By way of yet another embodiment, the above system is capable of end-to-end encryption for data transmission to and from end-users, customers, service providers, and between storage, optimization, training, execution and communications servers.

FIGURE 6 is a process flow diagram representing an embodiment of the development and training of an individual end-user infection prediction system utilizing a neural network. In this embodiment 600, physiological temperature data is measured from a self-powered device containing a processor, memory, Bluetooth communication system and a temperature transducer 610. This device is pre-configured for an individual end-user to automatically measure skin temperature at regular repeated intervals, over an extended period. Further, the device is configured to securely transmit time-based physiological data at regular intervals, according to set protocols to preconfigured hubs. Once data has been received by a Bluetooth hub 620, this hub transmits the data through a secured, wired ethemet network and is stored on a data storage server, where a database is generated of raw data 630. This database comprises time domain historical temperature data, where data generates a self-referential table for the subject, comprising skin temperature, time and date stored as sequential rows for each measurement period. Updated skin temperature data is added to the database as it is received on the server 630. A processor within the server manipulates data according to a specified protocol to enhance its structure. During this process 640, a reduction in dimensionality and input transformations are performed on the structure, followed by data compression. At this time, data may also be removed for any periods the device was not collecting user temperatures, or for times when the device was worn incorrectly. Once complete, a processed database 640 is enhanced for training machine learning and deep learning systems and saved to memory.

In this embodiment, the processed database is then transmitted through a secured connection to a cloud-based server specialized for generating machine learning models 650. A methodology (not shown) is then applied for evaluating sufficient data for training predictive end-user models.

After an enhanced dataset has been prepared and sufficient data collected, an individual end-user model is then trained. Development of models may be through traditional statistical techniques, non-limiting examples include linearized, mixed, k- nearest-neighbors (knn), Support Vector Machines (SVM) or Forrest models and the like. Non-limiting examples of more advanced techniques include deep learning networks, reinforcement learning networks and hybrid networks. Non-limiting examples of deep learning neural networks designed to evaluate time domain and sequence data include Recurrent Neural Networks (RNN) and Long-Short-Term-Memory networks (LSTM), variants of RNN. Convolutional Neural Networks (CNN) or hybrid networks, may also be used if sequencing input transformations have been applied to input data. Models take the form of having at least one input variable of skin temperature or heart beat data, with data collected over a sufficient period to inform an output variable that predicts abnormal rises in deep body temperature or abnormal body physiology.

In the embodiment illustrated in FIGURE 6, unsupervised deep learning neural network models are trained 660 using the processed database on a multi-processor, multithreaded server with specialized TPU's for training highly parallel models. This server may be equipped with sufficient RAM capable of loading complete datasets and model structures into active memory, and server architecture is specific to evaluating deep learning models on temporal and sequence datasets. In the embodiment shown in FIGURE 6, the dataset includes time of day and normal working (shift) times to ensure circadian and diurnal cycle variations are correctly tracked. Outputs from models provide an end-user probability of infection for each time period. After model training is complete, model layers with specific weights and biases are compiled 670. In this example, predictive outputs are validated 680 using a portion of the end-users processed dataset that was withheld prior to training. Once sufficient data has been collected to train and validate a model, this compiled network is passed back to the local server 630 through a secure connection. After the local server has activated this user's predictive system, real-time end-user tracking for that specific individual can begin 690. For this embodiment, during tracking, determinations of body temperature normality and probabilities of infection are evaluated in real-time on non-trained temporal subject temperature data. With reference again to FIGURE 4, current and historical data may be accessed and displayed through a connected portal as illustrated in FIGURE 4, where analytics are conducted on predictive outputs 495. In the embodiment illustrated in FIGURE 4, when probability of physiological abnormality increases above a predetermined level 470, results are identified as abnormal and a user(s) is classified as infected 480. In this embodiment, notification alerts are sent to the end-user in the form of an SMS and an email 491. For the embodiment illustrated in FIGURE 4, after a decision-making algorithm has been trained, the complete process from heart rate measurement to prediction and alerting of infection may be performed quickly, such as 5 minutes or less.

Development and training of infection predictions may be improved using supplementary data. Another embodiment of the process shown in FIGURE 6, as highlighted inside regions enclosed by 600 and 695, adds user infection test results, and applies these as an input which may then be used to train a supervised deep learning model, to identify and predict classification states of infected or normal.

Additional physiological, physical, or environmental data may also improve accuracy of predictions. FIGURE 7 shows examples of measured parameter enhancements that may be applied to improve predictive accuracy and confidence of embodiments of the disclosure. Additional measurements or data that may enable improvements are broadly split into five types, body temperature 710, temporal 720, other (non-temperature) body physiology 730, personal 740, and additional data 750. Specific examples of temperature measurements in addition to a single skin temperature site 711, include measuring skin temperatures of multiple body regions simultaneously 712. These may provide several enhancements, increasing the accuracy of body temperature estimates.

Inclusion of a non-vasoconstricting site with a vasoconstricting region can take advantage of the phenomena described above during a febrile response initiated through an infection. During this time, resetting of thermal effector zones attempts to attain higher body temperatures. To enable higher deep body temperatures triggered by fevers, multiple physiological changes occur. These may include increased basal metabolism, often created by initiation of shivering 734, peripheral vasoconstriction and/or reduction of sweating 733 to reduce evaporative heat loss. Peripheral vasoconstriction causes skin temperature reductions in those regions. Predictions may be enhanced by measuring skin temperature differences between skin regions that positively correlate with deep body temperatures i.e. no, or minor deep tissue vasoconstriction, and those regions that may negatively correlate with deep body temperature during a febrile response, for example hands and fingers. Comparing results of an infected end-user to that user's afebrile historical data, may provide predictions of abnormal body temperatures and therefore indicate an infection.

Additionally, some deployments may measure deep body temperature 713 directly, for example rectal, vaginal, oral, aural, axilla, stomach (through ingested pills) or esophageal temperatures. Changing periodicities of measurement time intervals 721 may affect prediction accuracies, as may altering overall model training periods 722. Increases in predictive accuracy rates may be achieved through simultaneously measuring user skin temperature and activity state through heart rate or beat rhythm 731, or user step rate 732. As described above, sweating activity 733 and shivering response 734 are both known to influence infection. Sweating activity can be measured by many methods, non-limiting examples include galvanic skin response and skin relative humidity.

Inclusion of end-user sex 741, may allow development of improved accuracy predictions which capture changes due to menstrual cycles in females. In addition to sex 741, other personal data 742, such as race, age, previous medical conditions, fitness, body mass index or birth control methods and the like, may provide statistical blocking inputs for further improvements to system predictive accuracy.

Additional physiological, environmental, and physical measurements may also increase accuracy of predicting physiological normality. Non-limiting examples of these measurement types include, end-user respiratory rate 751, blood oxygen saturation (SpCh) 752, cardiac health, in which a non-limiting example includes multi-lead electrocardiography (ECG) 753 or body movement 754. As discussed above, pathogenic type infections may affect efficiencies of critical organs such as the heart and lungs. Consequent effects of lower organ efficiencies may result in disturbed heart rhythms or in the case of lungs, lower oxygen uptake as measured by an SpCh test 752. In addition to measuring skin temperature directly, anon-limiting example of an alternative that measures physiology related to vasculature is peripheral skin perfusion 755. Environmental factors such as ambient temperature, absolute or Relative Humidity (RH) 756, geographic location 757 measured through GPS or other means, and ambient lighting conditions 758, may all influence body temperature and therefore provide improved predictive accuracy .

Use of standard mathematical tools can enhance decision-making algorithm development, through screening-training that employs techniques for manually or automatically identifying dimensionality reduction and removing redundant variables that provide little or no improvement to predictive accuracies 760. To increase predictive accuracies of end-user infection tracking, models may also be periodically or constantly retrained 770 and updated 771. Both predictive and recent historical tracking records can be evaluated and integrated to provide new parameters and data structures 780, non-limiting examples include infection rates, geography's and epidemiology's of local populations, workplaces, teams, and the like.

While developing systems for single end-users enables predictive capabilities for a specific individual, the usefulness of this disclosure is enhanced through training master models with populations of end-users. For this disclosure, a population is defined as more than one individual. FIGURE 8 shows one example of this type of implementation 800. This embodiment applies the methods described in FIGURE 6 to a dataset(s) comprised of a population(s) of end-user's data. Measurements of a population's physiological, physical, or environmental data can be measured simultaneously or over different time periods. In this embodiment, population skin temperature data 610, is recorded and transmitted to Bluetooth hubs 620, using similar physiological measurement devices and hubs described for embodiments in FIGURE 6. Population data is collected and stored in a relational database hosted on a cloud-based server 830.

In this embodiment, a population database of measurements comprises time domain historical temperature data, where the data generates self-referential tables for each subpopulation e.g. each workplace and team. Additionally, rows in the database include a subject specific tag unique to that end-user. Multiple users are contained within each table and similar to single user embodiments, skin temperature, time and date are recorded as sequential rows within a relational database table. Skin temperature data from each user is measured at regular intervals over an extended period. Within a population, actual time of an individual measurement and intervals between measurements, may vary across endusers. A processor within the data server manipulates data according to a specified protocol, blocking by individual end-user for structure enhancement. During this enhancement process, where each end-user's data is treated independently, the structured dataset undergoes input transformations, a reduction in dimensionality, followed by dataset compression according to a pre-defined protocol. Similar to the single end-user case in FIGURE 6 described above, data from individual users may also be removed for any period the device was not collecting user temperatures or when worn incorrectly. Once complete, a processed database of population data is enhanced for training machine learning and deep learning systems and saved to memory 840.

For embodiments illustrated in FIGURE 8, a protocol is applied to the processed database (not shown) to evaluate whether sufficient data has been collected within each user, that is, enough temperature data is available to compile a complete dataset for an individual user, and between users, that is, enough complete end-user datasets are available for training a predictive population model. In the embodiment illustrated in FIGURE 8, when sufficient data has been compiled in the processed database to meet these criteria, sub-population data is transmitted through a secured connection to a cloud-based server specialized for generating machine learning models 650. This embodiment uses the same hardware for training models as that described for single end user cases in FIGURE 6. For population model training, additional personal information may be included in datasets to improve the reliability and accuracies of predictions.

Multiple input variables, multidimensional blocks and large quantities of data make a population model an ideal candidate to be resolved using deep neural networks. Predictive outputs can be in regressive forms i.e. probability of temperature/physiology normality, or probability of infection; or a classification output, where an end-user is classified as normal (not infected) or abnormal (infected or febrile). For embodiments illustrated in FIGURE 8, variable selection 855 is dependent on the type of predictive model being trained.

In one embodiment shown in FIGURE 8, comprising regions 800 and 860, predictive outputs are in the form of probabilities of infection, where development and training of an unsupervised deep learning neural network model follow similar procedures described for single end-user cases provided in FIGURE 6 600. While similar in form, training population models may involve additional parameters, complexity, and computing time, due to multiple end-users being included within each dataset. By way of another embodiment, combining 800 and 870, a classification model is trained using supervised techniques 875. In this embodiment, additional model inputs of infection status and time period over which infections occur are known for each end-user 695 as described for the single end-user cases, and these are included to train a supervised model. Master population models may also be developed from hybrid or reinforcement types of machine learning. For embodiments illustrated in FIGURE 8, master population models are trained and underlying structures of layers, weights, and biases for both unsupervised and supervised models are compiled and saved 880. These compiled models are validated 890 using portions of measurement data that was withheld prior to training. In another embodiment (not shown), system predictions are continually validated and improved through the addition of more population datasets.

Transfer learning is a type of model training that allows new models to be trained quickly, in which the original model inputs, hidden structures (often called layers) and output structure are retained, while weights and biases are customized using data from specific end-users. For embodiments illustrated in FIGURE 8, master population models remain on the specialized machine learning server. By including additional data from existing users, and as new users are added 610, individual end-user predictive models are quickly developed through transfer training techniques 660 from master models, validated 680 and then passed back to the cloud server 830 for activation. By way of yet another embodiment, referring to FIGURE 6, a master population model 880 is passed back to the local server, which is then used to train individual end-user models 660 using transfer learning. These end-user models are validated 680 using portions of reserved end-user's data, and then compiled user specific models are stored for individual activation.

The methodology and apparatus of this disclosure can be employed within any type of work, educational, sports, military, recreational, or health space. A few non-limiting examples include high rise office buildings or complexes, manufacturing factories or warehouses, retail malls, schools, higher education facilities, professional sports teams and supporting staff, military bases, ships, hospitals, nursing homes, doctors' offices, fitness complexes, health centers, hotels, and the like.

FIGURE 9 shows a block-flow diagram representation for the tracking system of the embodiment illustrated in FIGURE 4. Embodiments shown in FIGURE 9 provide realtime predictions for early identification of pathogenic infections within workforces. Additionally, through removal of infected individuals soon after a fever occurs, this system can help prevent workplace epidemics. In this embodiment, a subscription-based model provides four users with a choice of measurement hardware 421 for measuring heart rate data. For convenience in the example, each user has selected a watch worn on the wrist. These heart rate measurement devices are pre-configured to provide a measurement from each user every 60 seconds, for example. After data is collected it is sent from the device through dedicated hubs 430 to the local server 440. Each user's device is pre-configured to first try a low energy Bluetooth connection for data transmission, and if this is not possible, the device attempts to use an IEEE 802.16 type wireless connection. If no connection to hubs are currently possible, measurement devices store data in local memory until such a time as a connection becomes available. In another embodiment (not shown), physiological measurement devices can incorporate loT transceivers and connect directly to cellular networks.

In the embodiment illustrated in FIGURE 9, after user heart rate data has been received by a hub 430, it is transmitted, via secure ethemet connection to the local server 440. On this server, raw data is preprocessed, adding a unique user identification tag, time, date, heart rate and the user's current location based upon the specific hub 430 that received data. Each user's data for a specific time period is then stored as a unique record in a relational SQL database. For this deployment, all four end-user decision making algorithms have been previously compiled, stored and activated on the local server 440. Algorithms used in this deployment are built with a sequential type deep learning network. A temporary input dataset 450 for an individual end-user evaluation, combines historical data with transformations to become the input for that model at the current time period. In one preferred embodiment, a new temporary dataset may be created within 30 seconds of the user record being written to memory.

Temporary dataset's 450 are evaluated by decision-making algorithm specific to individual user's 460. Output predictions inform decisions 470 in near real-time, such as within 5 minutes of physiological measurement data being received for each end-user. In this embodiment, predictive outputs provide classifications of user normal 475 or user infected 480, and the decision tree directs actions for each user based on the temporal output. If an end-user is determined as normal for that time period, their data record is updated 465, passed to the SQL database, the portal is updated 495, and no other action is taken. If an end-user is determined as infected 480, the local server can send real-time notification alerts 490 through e-mails, the internet, and send text messages or voice mails through a VoIP server, for example. In this embodiment, notification alerts take forms of text messages using Short Messaging Service (SMS) protocol, sent through a cellular service to the end-user's mobile device 491. An e-mail 491 may also (or alternatively) be sent to the end-user's registered e-mail address. Additionally, an e-mail 492 may be sent to a registered safety officer from the Human Resources (HR) department. E-mails are sent through an internet connection, and a flag 493, is sent to identify this event on the customer portal 495. In this embodiment, the portal 495 may be accessed by any device with correct credentials and an internet connection. That user's data record is updated 465 and passed to the SQL database, this update reflects a change in status and that notification alerts have been dispatched.

In one embodiment, when an end-user is identified as infected, that user removes themself from active duty and additional actions are taken before returning to the workforce. These actions can take many forms, non-limiting examples include medical tests, inoculation, or isolation periods. In another embodiment, end-users submit to a negative infection test result 695, before returning to the workforce. In this embodiment, test results are also stored on the local server 440 and end-user models are re-trained monthly using most recent datasets.

In another embodiment, workplaces equip employees with implementations of this disclosure and software portals provide epidemiological analytics on workforce populations, examples include demographics on frequency of cases, changes in rates of infection and the local facility and global geography of case development. By way of yet another embodiment, infection testing results are integrated into portal analytics and continue to improve epidemiological predictions through re-training with updated test data.

FIGURE 10A shows an exemplary dataset from an embodiment of this disclosure. For this embodiment, forehead skin temperature was measured for three end-users over 45 workdays, 63 days total. Skin temperature was sampled for each user at 30-minute intervals during time spent in an office workplace. The 'x' axis displays date of measurements, while the primary 'y' axis (left) shows forehead skin temperature of three end-users. The secondary 'y' axis (right), records subject confidence of a decision-making tracking system. All three individuals had a typical workday of ~8 hours from Monday to Friday. Measured temperature for each user varies throughout the workday and timings of temperature measurement intervals were not synchronized across users. Measurement devices were equipped to identify if they were being worn, during dormant periods the devices recorded a temperature of 0°C. Referring to FIGURE 10A, it is observed that each of the three individuals had different recorded mean temperatures, with User 2 providing the lowest mean of ~34°C, User 3 providing a mean of ~35°C, and User 1 a mean of ~36°C. These temperature differences may be due to actual differences in body temperature, or to measurement device usage and calibration errors. During the measurement period of early January to early March, there were two confirmed cases of fever caused by a flu type infection. One of these periods 1030 is observed for User 2 during the fifth week of tracking, the second was during the last two measurement days for User 3 1040.

Two deep learning models were transfer trained for each of these three individual users using trained neural networks from larger populations. Two models were trained for each user to compare predictive accuracies of systems that use a regression basis, to those using a classification basis. During the training period, confidence associated with each model output, subject confidence 1010, was calculated according to a pre-determined protocol in near real-time, as temperature data was added to the datasets. For this embodiment, a pre-determined minimum confidence level of 99% was set to identify end of training and that tracking could begin. Additionally, according to this protocol, both models reached this level before user tracking could start. For this embodiment, that level was reached 10 measurement days after physiological monitoring began, as illustrated by the vertical dashed line 1020. To the right of line 1020, each end-user was now actively tracked in real-time while at their place of employment.

FIGURES 10B and 10C illustrate examples of systems based on regression modelling to track and predict thermal fevers. The figures display data for both individuals who became febrile in FIGURE 10A. This tracking system was applied to each of these users for 35 workdays (periods to the right of line 1020). The 'x' and primary 'y' axes of these two figures are the same as FIGURE 10A. The secondary 'y' axes (right) of these figures, displays probability of infection for the regression-based tracking system. As discussed for FIGURE 10A, periods to the left of line 1020 were used for system training.

FIGURE 10B shows results of a tracking system for User 2, while FIGURE 10C shows results of a tracking system for User 3. Probability of infection prediction for FIGURE 10B is calculated from User 2's skin temperature data, while probability for FIGURE 10C is from User 3's data. Tracked probabilities of infection from regression models vary over time for each user, according to individual weights and biases. In this non-limiting example, horizontal dotted line 1050, signifies a 50% threshold limit for alerts to be sent to end-users and customers. When probabilities are based upon regression models, time is required for probabilities of infection to rise from a value below the threshold to one above. In both cases ~1 hour (1 - 2 time intervals) passed between the time prediction probability rises above the noise floor of -10% to a value above the 50% cutoff threshold 1050. Additionally, in both cases, penetration of this cutoff threshold occurs during the first two hours of the first measured febrile day, as observed when comparing against temperature data 1030 and 1040.

FIGURES 10D and 10E show examples of systems based on classification models, to track and predict thermal fevers for the same two individuals. The 'x' and 'y ' axes are the same as FIGURE 10A. For systems with a binary classification basis (2 classes), no additional axis is required, as outputs are classified as either user normal or user infected, where predictions of user infected are illustrated by single hatched shaded regions. As for previous figures, FIGURE 10D shows a tracking system for User 2, while FIGURE 10E displays a tracking system for User 3. Once again, infection predictions for FIGURE 10D are calculated from User 2's skin temperature data, while predictions for FIGURE 10E are calculated from User 3's data. Tracked predictions from these classification models are binary, and therefore transition from user normal to user infected and vice versa occurs within a single period. In this non-limiting example, transitions in predictions to user infected, occurred during the first two hours (4-time intervals) of the first febrile measured day for both users, as observed when compared against temperature data for User 2 1030 and User 3 1040.

Typically, predictive results from embodiments of this disclosure, achieve accuracies of fever tracking and prediction as measured using precision (PPV) of -95% within the first measurement day after a user becomes febrile.

Therefore, in one embodiment, the disclosure provides an apparatus configured to provide a means of tracking and characterization of subject body temperatures as normal or abnormal, wherein the system comprises: (a) an apparatus for measuring at least one body temperature of subjects, wherein the temperature transducer is configured to be powered without tethering to a specific location; the transducer automatically measures body temperature on a repeated basis and transducer circuitry is interval configurable by those of normal skill in the art; (b) the transducer is in communication with a processor connected to a storage system, wherein data from the processor is transmitted to a computer according to a protocol and data from transducers may be viewed on a display device in near real-time; (c) a database is generated comprising at least one unique subject identification tag, one body temperature, time and date for each measurement, recorded over at least two time periods; (d) a processed dataset is generated and stored for use by a predefined training protocol, wherein the dataset of all subject measurements is connected to a processor for data reduction and structure optimization by dimensionality reduction, input transformations and data compression, wherein the dataset is saved to memory and storage enhanced for training models; (e) a methodology for evaluating sufficient data for training a predictive population system; (f) a model is created, wherein the processed database is passed to a processor(s), the processor(s) employs a protocol to train a customized model, with architecture specific to evaluating temporal and sequence datasets; (g) a characterization and tracking system is activated and evaluated to make determinations of body temperature normality on non-trained temporal subj ect temperature data, wherein the trained model comprises at least one predictive temporal output of each subject's temperature normality with quantified prediction uncertainty, to enable corrective actions and provide at least one notification to a customer or end-user.

In one embodiment, systems are trained using data created from a database of historical or simulated computer temperature data. In another embodiment, the system's prediction of body temperature normality is based on using multiple inputs from at least one body temperature and at least one additional input of body physiology, thermal state, activity, health, physical state, or environment. In one embodiment, prediction of body temperature normality is based on correlations using more than one body temperatures at more than one time points. In another embodiment, the system's prediction of the subject's body temperature normality is based on correlations of one input from a region with no, or inconsequential deep tissue vasoconstriction and a second input of the difference between the first input and a temperature from a peripheral body vasoconstricting region. In one embodiment, a new dataset and model is trained for each subject, wherein the new model comprises at least one input of body temperature measured over an extended period for each subject and provides at least one predictive temporal output of each subject's body temperature normality with quantified prediction uncertainty. In another embodiment, probability of infection is predicted based on predicted normality of body temperature. In one embodiment, an individual end-user model is transfer trained from a population model that has been trained on a different dataset, wherein a new model comprises at least one input of body temperature measured over an extended period from one or more subjects and provides at least one output of each subject's prediction of infection with quantified prediction uncertainty. In another embodiment, prediction of infection probability is improved based on re-training neural model(s) with more recent measurement data from a subject(s). In another embodiment, prediction of infection probability is further classified into at least two classes of not infected or infected. In yet another embodiment, a processor is configured to continually make infection predictions on real-time or near real-time temperature measurement data.

In another aspect, the disclosure provides an apparatus configured to provide a means of tracking and characterization of subject body physiology as normal or abnormal, wherein the system comprises: (a) an apparatus for measuring heart rate in beats per second (BPS) or heart beat rhythm of subjects, wherein the heart beat transducer is configured to be powered without tethering to a specific location; the transducer automatically measures human heart beat data on a repeated basis and transducer circuitry is interval configurable by those of normal skill in the art; (b) the transducer is in communication with a processor connected to a storage system, wherein data from the processor is transmitted to a computer according to a protocol and data from transducers may be viewed on a display device in near real-time; (c) a database is generated comprising historical data, wherein the historical data generates a self-referential table for subjects comprising at least one unique subject identification tag, one heart rate in BPS or heart beat rhythm, time and date for each measurement, recorded over at least two time periods; (d) a processed dataset is generated and stored for use by a predefined training protocol, wherein the dataset of all subject measurements is connected to a processor for data reduction and structure optimization by dimensionality reduction, input transformations and data compression, wherein the dataset is saved to memory and storage enhanced for training models; (e) a methodology for evaluating sufficient data for training a predictive population system; (I) a model is created, wherein the processed database is passed to a processor(s), the processor(s) employs a protocol to train a customized model, with architecture specific to evaluating temporal and sequence datasets; (g) a characterization and tracking system is activated and evaluated to make determinations of physiology normality on non-trained temporal subject heart beat data, wherein the trained model comprises at least one predictive temporal output of each subject's physiology normality with quantified prediction uncertainty, to enable corrective actions and provide at least one notification to a customer or end-user.

In one embodiment, the system is trained using data created from a database of historical or simulated computer heart beat data. In another embodiment, the system's prediction of body physiology normality is based on using multiple inputs from a heart rate in BPS, or heart beat rhythm and at least one additional input of body physiology, thermal state, activity, health, physical state, or environment. In one embodiment, the system's prediction of body physiology normality is based on correlations using heart beat data measurement and at least one body temperature at more than one time point. In another embodiment, a new dataset and model is trained for each subject, wherein the new model comprises at least one input of heart beat data measured over an extended period for each subject and provides at least one predictive temporal output of each subject's body physiology normality with quantified prediction uncertainty.

In one embodiment, probability of infection is predicted based on predicted normality of body physiology. In another embodiment, an individual end-user model is transfer trained from a population model that has been trained on a different dataset, wherein the new model comprises at least one input of heart beat data measured over an extended period from one or more subjects and provides at least one output of each subject's prediction of infection with quantified prediction uncertainty. In one embodiment, prediction of infection probability is improved based on re-training neural model(s) with more recent measurement data from a subject(s). In another embodiment, prediction of infection probability is further classified into at least two classes of not infected or infected. In another embodiment, a processor is configured to continually make infection predictions on real-time or near real-time heart beat measurement data. In yet another embodiment, a processor connected to a display and communication system is configured to display in real-time or near real-time infection results of tracked subjects and through communication hardware provides near-real-time reporting and notifications to customers/end-users of infection condition.

In one aspect, the disclosure provides a method of detecting a disease, comprising using the system of the disclosure. In one embodiment, the disease is caused by an infection. In another embodiment, the disease is coronavirus Disease 2019 (COVID-19).

In one aspect, the disclosure provides a method of predicting infection epidemiology, comprising using the system of the disclosure.

Other embodiments include combinations and sub-combinations of features described or shown in the drawings herein, including for example, embodiments that are equivalent to: providing or applying a feature in a different order than in a described embodiment, extracting an individual feature from one embodiment and inserting such feature into another embodiment; removing one or more features from an embodiment; or both removing one or more features from an embodiment and adding one or more features extracted from one or more other embodiments, while providing the advantages of the features incorporated in such combinations and sub-combinations. As used in this paragraph, feature or features can refer to the structures and/or functions of an apparatus, article of manufacture or system, and/or the steps, acts, or modalities of a method.

Various embodiments are described above with reference to block diagrams and/or flowchart illustrations of apparatuses, methods, systems, and/or computer program instructions or program products. It should be understood that each block of any of the block diagrams and/or flowchart illustrations, respectively, of portions thereof, may be implemented in part by computer program instructions, e.g., as logical steps or operations executing on one or more computing devices. These computer program instructions may take the form of applications, programs, program modules, scripts, source code, program code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like (also referred to herein as executable instructions, instructions for execution, program code, computer program instructions, and/or similar terms used herein interchangeably).

These computer program instructions may be loaded onto one or more computers or computing devices, such as special purpose computer(s) or computing device(s), or other programmable data processing apparatus(es) to produce a specifically-configured machine, such that the instructions which execute on one or more computer or computing devices or other programmable data processing apparatus provide operations for or implement the functions specified in the flowchart block or blocks and/or carry out the methods described herein.

These computer program instructions may also be stored in one or more computer- readable memory or portions thereof, such as the computer-readable storage media, that can direct one or more computers or computing devices or other programmable data processing apparatus(es) to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including computer- readable instructions for implementing the functionality specified in the flowchart block or blocks.

It will be appreciated that the term computer or computing device can include, for example, any computing device or processing structure, including but not limited to a processor (e.g., a microprocessor), a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a system on a chip (SoC), or the like, or any combinations thereof. Accordingly, blocks of the block diagrams and/or flowchart illustrations support various combinations for performing the specified functions, combinations of operations for performing the specified functions and program instructions for performing the specified functions. Again, it should also be understood that each block of the block diagrams and flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, or portions thereof, could be implemented by special purpose hardware-based computer systems or circuits, etc., that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

According to some embodiments, many individual steps of a process may or may not be carried out utilizing computer or computing based systems described herein, and the degree of computer implementation may vary, as may be desirable and/or beneficial for one or more particular applications.

In the foregoing description, specific details are set forth to provide a thorough understanding of representative embodiments of the present disclosure. It will be apparent to one skilled in the art, however, that the embodiments disclosed herein may be practiced without embodying all of the specific details. In some instances, well-known process steps have not been described in detail in order not to unnecessarily obscure various aspects of the present disclosure. Further, it will be appreciated that embodiments of the present disclosure may employ any combination of features described herein.

The present application may reference quantities and numbers. Unless specifically stated, such quantities and numbers are not to be considered restrictive, but exemplary of the possible quantities or numbers associated with the present application. Also in this regard, the present application may use the term "plurality" to reference a quantity or number. In this regard, the term "plurality" is meant to be any number that is more than one, for example, two, three, four, five, etc. The terms "about," "approximately," "near," etc., mean plus or minus 5% of the stated value. For the purposes of the present disclosure, the phrase "at least one of A and B" is equivalent to "A and/or B" or vice versa, namely "A" alone, "B" alone or "A and B.". Similarly, the phrase "at least one of A, B, and C," for example, means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C), including all further possible permutations when greater than three elements are listed.

Throughout this specification, terms of art may be used. These terms are to take on their ordinary meaning in the art from which they come, unless specifically defined herein or the context of their use would clearly suggest otherwise. The principles, representative embodiments, and modes of operation of the present disclosure have been described in the foregoing description. However, aspects of the present disclosure which are intended to be protected are not to be construed as limited to the particular embodiments disclosed. Further, the embodiments described herein are to be regarded as illustrative rather than restrictive. It will be appreciated that variations and changes may be made by others, and equivalents employed, without departing from the spirit of the present disclosure. Accordingly, it is expressly intended that all such variations, changes, and equivalents fall within the spirit and scope of the present disclosure, as claimed.

Claims

CLAIMS The embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. A system to track and characterize subject body temperatures, the system comprising: a temperature transducer to measure at least one body temperature of a subject, wherein the temperature transducer is coupled to a mobile power source, the transducer being configured to measure body temperature on a repeated basis and the temperature transducer circuitry is interval configurable; a processor in operative communication with the temperature transducer, the processor being connected to a local storage system, wherein data received from the temperature transducer is transmitted to one or more remote computer(s) according to a communication protocol; a remote database resident on the one or more remote computer(s), the remote database being programmed to store historical data for a plurality of subjects, wherein the historical data for each of the plurality of subjects comprises at least one unique subject identification tag that uniquely identifies a corresponding subject, at least one body temperature measurement for the corresponding subj ect, and temporal information for each temperature measurement; wherein the one or more remote computer(s) is (are) configured to create a processed dataset from the historical data, the processed dataset comprising a subset of the historical data that is enhanced for evaluation by a training methodology; wherein the one or more remote computer(s) is (are) further configured to implement the training methodology on the processed dataset, the training methodology operative to create a statistical model based on a sequence of subject temperatures for each subject having corresponding data in the remote database; wherein the one or more remote computer(s) is (are) still further configured to implement a characterization methodology to evaluate body temperature normality on nontrained temporal subject temperature data, wherein the characterization methodology comprises at least one predictive temporal output of each subject's temperature normality with quantified prediction uncertainty, to enable corrective actions and provide at least one notification to a customer or end-user.

35

2. The system of claim 1, wherein the enhancement of the historical data comprises at least one or more of data reduction, structure enhancement by dimensionality reduction, input transformations, and data compression.

3. The system of claims 1 or 2, wherein training data is created from a database of temperature measurements from individuals other than the plurality of subjects or simulated computer temperature data.

4. The system of claims 1-3, wherein prediction of body temperature normality is based on using multiple inputs from at least one body temperature and at least one additional input of body physiology, blood oxygen level, thermal state, activity, health, physical state, or environment.

5. The system of claims 1-4, wherein prediction of the subject's body temperature normality is based on correlations using more than one body temperature at more than one time point.

6. The system of claim 1 -5, wherein the processor in operative communication with the temperature transducer is further connected to a display device, and further wherein the at least one body temperature is viewable in real time or near real time on the display device.

7. The system of claim 1-6, wherein a new dataset and model is trained for each subject, wherein the new model comprises at least one input of body temperature measured over an extended period for each subject and provides at least one predictive temporal output of each subject's body temperature normality with quantified prediction uncertainty.

8. The system of claim 1-7, wherein a probability of infection is predicted based on predicted normality of body temperature.

9. The system of claim 8, wherein an individual end-user model is transfer trained from a population model that has been trained on a different dataset, wherein the individual end-user model comprises at least one input of body temperature measured over

36 an extended period from one or more subjects and provides at least one output of each subject's prediction of infection with quantified prediction uncertainty.

10. The system of claim 8, wherein prediction of infection probability is improved based on re-training neural model(s) with more recent measurement data from a subject(s).

11. The system of claim 8, wherein prediction of infection probability is further classified into at least two classes of not infected or infected.

12. The system of claim 11, wherein the system is configured to continually make infection predictions on real-time or near real-time temperature measurement data.

13. A system to track and characterize subject body temperatures, the system comprising: a heart sensor to measure a heart rate of a corresponding subject, wherein the heart sensor is coupled to a mobile power source, the heart sensor being configured to measure at least heart rate on a repeated basis and the heart sensor circuitry is interval configurable; a processor in operative communication with the heart sensor, the processor being connected to a local storage system, wherein data from the heart sensor is transmitted to one or more remote computer(s) according to a communication protocol; a remote database resident on the one or more remote computer(s), the remote database being programmed to store historical data for a plurality of subjects, wherein the historical data for each subject comprises at least one unique subject identification tag that uniquely identifies a corresponding subject, at least one heart rate measurement for the corresponding subject, and temporal information for each heart rate measurement; wherein the one or more remote computer(s) is (are) configured to create a processed dataset from the historical data, the processed dataset comprising a subset of the historical data that is enhanced for evaluation by a training methodology; wherein the one or more remote computer(s) is (are) further configured to implement the training methodology on the processed dataset, the training methodology operative to create a statistical model based on a sequence of subject heart rates for each subject having corresponding data in the remote database; wherein the one or more remote computer(s) is (are) still further configured to implement a characterization methodology to evaluate heart rate normality on non-trained temporal subject heart rate data, wherein the characterization methodology comprises at least one predictive temporal output of each subject's heart rate normality with quantified prediction uncertainty, to enable corrective actions and provide at least one notification to a customer or end-user.

14. The system of claim 13, wherein training data is created from a database of heart rate measurements from individuals other than the plurality of subjects or simulated computer heart rate data.

15. The system of claims 13-14, wherein prediction of body physiology normality is based on using multiple inputs from a heart rate in BPS, or heartbeat rhythm and at least one additional input selected from a group comprising body physiology, blood oxygen level, thermal state, activity, health, physical state, and environment.

16. The system of claims 13-15, wherein prediction of the subject's body physiology normality is based on correlations using heartbeat data measurement and at least one body temperature at more than one time point.

17. The system of claim 13-16, wherein a new dataset and model is trained for each subject, wherein the new model comprises at least one input of heart beat measured over an extended period for each subject and provides at least one predictive temporal output of each subject's body physiology normality with quantified prediction uncertainty.

18. The system of claim 13-17, wherein probability of infection is predicted based on predicted normality of body physiology.

19. The system of claim 18, wherein an individual end-user model is transfer trained from a population model that has been trained on a different dataset, wherein the new model comprises at least one input of heartbeat data measured over an extended period from one or more subjects and provides at least one output of each subject's prediction of infection with quantified prediction uncertainty.

20. The system of claim 18, wherein prediction of infection probability is improved based on re-training neural model(s) with more recent measurement data from one or more subject(s).

21. The system of claim 18, wherein prediction of infection probability is further classified into at least two classes of not infected or infected.

22. The system of claim 20, wherein the system is configured to continually make infection predictions on real-time or near real-time heartbeat measurement data.

23. The system of claim 13-17, wherein the heart rate data comprises a heart rhythm.

24. The system of claim 1-12, wherein another processor connected to the one or more remote computer(s), to a display, and to a communication system is configured to display in real-time or near real-time, infection results of tracked subjects and to provide near-real-time reporting and notification to customers/end-users of infection condition.

25. A method of detecting a disease using any one of the systems of claims 1-23.

26. The method of claim 25, wherein the disease is caused by an infection.

27. The method of claim 25, wherein the disease is a coronavirus disease.

28. The method of claim 27, wherein the coronavirus disease is COVID-19.

29. A method of predicting infection epidemiology using any one of the systems of claims 1-23.

39