CN101526465B - Quick multi-wavelength tissue optical parameter measuring device and trans-construction method - Google Patents

Abstract

The invention belongs to the field of optical parameter measurement in tissue optics research, and relates to a multi-wavelength tissue optical parameter trans-construction method based on a neural network. The method comprises: using double integrating sphere technology and a high-sensitivity photoelectric detector to collect reflected light and transmitted light on the surface of a tested sample, and using a method of relative measurement to obtain the reflectivity, diffuse transmittance and collimation transmittance of the sample; using a Monte Carlo method to establish a mapping model of a tissue optical parameter (mua, mus and g) to a measurement amount (Rd, Td and Tc); establishing a BP neural network containing two buried layers, and adopting a gradient descent weight learning function to train a BP network; and predicting the optical parameter of the measured sample through the (Rd, Td and Tc) obtained by measurement in real time, wherein an input signal of the neural network is (Rd, Td and Tc), and an output signal is (mua, mus and g). The invention provides a device adopted by the method at the same time. The method provided by the invention can quickly and accurately measure the tissue optical parameters under multi-wavelength.

Description

Quick multi-wavelength tissue optical parameter measuring device and trans-construction method

Technical field

The invention belongs to the optical parameter measurement field in the organism optical research, be specifically related to a kind of multi-wavelength in vitro tissue optical parametric detection method.

Background technology

Along with the progress of society, the mankind more and more pay attention to the health of self, are also paid attention to by people gradually with the healthy relevant detection means and the method for early diagnosis and diagnosis and treatment.The detection of optical correlation and diagnosis and treatment means possess advantages such as high resolving power, multiparameter and high time-space resolution, have Wicresoft again even do not have the advantage of wound, no ionization aspect medical diagnosis.Therefore, optical correlation method having wide practical use in medical diagnosis and treatment.

Biomedical Photonics is the subject of using in biology, medical domain about light.Find X ray from the roentgen, optics just constantly is applied in the medical diagnosis on disease aspect, along with the continuous development of optical technology, also provides new means for biology, medical science.The eighties in 20th century, people begin to attempt utilizing the optical correlation technology that human body is not had wound and diagnosis detects in Wicresoft.Developed country has also given the attention of height and has constantly strengthened the scientific research input the research of Biomedical Photonics.The symposial that America NI H holds thinks that the method for Biomedical Photonics can be used for the early detection of cancer and other diseases, diagnosis and treatment.A new generation is also just turning to the cancer clinical practice at the bulk optics imaging technique from the laboratory.The delivering of paper on European Biomedical Photonics meeting of holding every year and the internal authority publication proves that also the photonics technology bringing into play crucial effects.

Organism optical is the research propagating characteristic of light in biological tissue's body and the subject of rule in the Biomedical Photonics, be the basic theory of medical science photon technology, with light in biological tissue's body kinematics (as the propagation of light) problem and dynamics (as the detection of light) problem as mainly studying a question.Organizer is from molecular level, and most important chemical molecular is biomacromolecule (protein, nucleic acid, carbohydrate, lipid), water, vitamin and trace element in the constituent.Absorbing material in the organizer is mainly the cytochrome that all exist in pigment in haemoglobin in water, the blood, blood sugar, the skin, myoglobin, lipid and each cell in the muscle in ultraviolet to infrared band ^[1]And the macromolecular substances that contained in the tissue, protein for example, fat, mitochondrias etc. then have strong scattering process to photon.When the amount of institutional framework in the human body and various compositions changes, can cause that just the optical characteristics in the organizer changes.Promptly when in the human body certain physiology and pathological change taking place, can cause the variation of light travel path in organizer accordingly.For example, when canceration took place in organizer, the volume of blood flow of cancerous issue part can increase, and corresponding variation can cause the increase of absorption coefficient, so the variation of absorption coefficient is as the foundation of early diagnosis of cancer.Therefore, light is communication process and the physiology of path and organizer, the close ties of pathological state in organizer, become the emphasis that people study concern.

Tissue optical parameter is a series of physical quantitys that are used for describing light propagation state in organizer.By mensuration to each optical property parameter in the organizer, just can be indirect obtain in the organizer corresponding physiology, pathological information.Therefore the optical parametric of organizer has great importance for the detection and the early diagnosis of photodynamics detection and clinical diagnosis and disease.The interactional several performance patterns and the phenomenon of organizer and light mainly comprise: absorb, and scattering, reflection, refraction, luminous, photochemistry, optoacoustic etc.Corresponding optical parametric of drawing organizer mainly comprises:

1) scattering coefficient μ _s: be used for the long last probability that photon is scattered of the unit of description journey.Scattering coefficient is the function of wavelength, is mainly used to reflect the size of organizer to the scattering of light degree.

${\mathrm{\μ}}_s=\frac{\mathrm{dI}}{I}/\mathrm{dx}$

Dx in the formula---unit path

DI---photon is because of being scattered the luminous energy of losing in the unit path

The energy of photon before I---scattering events takes place.

The scattering of light in organizer mainly is different the causing of refractive index by micro-structure in the biological tissue (as nucleus and collagenous fibres) edge.When pathological change takes place in biological tissue, just changing along with tissue property with body.For example, when tissue canceration, the thickness of nuclear size and epithelial layer all can increase, and as the cell quantity increase of scatterer, thereby makes that the light scattering characteristic of normal structure and cancerous issue is different.

2) absorption coefficient μ _a: be used for the long last absorbed probability of photon of the unit of description journey.

${\mathrm{\μ}}_a=\frac{\mathrm{dI}}{I}/\mathrm{dx}$

Dx in the formula---unit path

DI---photon is because of being absorbed the luminous energy of losing in the unit path

The energy of photon before I---absorption events takes place

Why organizer has absorption is that chromophore is the chemical group that can cause carrying out in the spectral region interior focusing of 200～1000nm selective absorption in the molecule because it has chromophore.At more near infrared zones and infrared region, water is prevailing absorbing material in biological tissue's body.At the visible region of wavelength less than 600nm, the absorption meeting of light is owing to the influence of haemoglobin, melanin and other pigments raises.At ultraviolet band, again can be owing to protein, the strong absorption of nucleic acid and raising.In infrared band (600-1300nm), the low absorption appears in water, and therefore, this wave band is called as the optical window of near-infrared measuring and treatment.Concerning most of soft tissues, light absorption is relatively low, and scattering phase is to stronger.

3) anisotropic factor g: the cosine weighting mean value of scattering angle when being defined as scattering is called average scattering cosine again.The anisotropic factor has been pointed out the directionality of organizer's scattering.Can express the size of photodistributed heterogeneity in the organizer or forward scattering simply, intuitively.

$g=<\cos \mathrm{\&theta;}>=\frac{\&Integral;p\left(\mathrm{\&theta;}\right)\cos \mathrm{\&theta;d}\cos \mathrm{\&theta;}}{\&Integral;p\left(\mathrm{\&theta;}\right)d\cos \mathrm{\&theta;}}$

For animal tissue, g is generally 0.68～0.96.

4) refractive index n: when light non-normal incidence to two kind of different medium was surperficial, light path can change, and refractive index is exactly to describe the physical quantity that medium changes light path ability size.

$n=\frac{c_0}{c}$

c ₀Be light velocity of propagation in a vacuum; C is the velocity of propagation of light in medium.

For tissue, photonic absorption and scattering determined the space distribution situation of light in tissue, but at boundary, the influence of the refractive index of tissue is the most remarkable, and the scattering organizer in is discontinuous the causing of refractive index owing to organizer itself.Therefore, the refractive index of tissue, the conduction in tissue has certain influence to photon.

5) promote the parameter that obtains by above-mentioned each parameter:

Total attenuation coefficient: μ _t=μ _s+ μ _a, being used for describing simultaneously absorption and scattering effect, its characterizes the probability index that light is decayed in tissue, exist the media of absorption and scattering to be called as the chaos medium simultaneously;

Mean free path

When light is propagated in tissue, photon once absorb and the scattering generation before the mean distance of being passed by be called mean free path;

Reduced scattering coefficient μ _s': μ _s'=μ _s(1-g).

The optical parameter measurement of biological tissue can be divided into and exsomatize measuring and in two kinds of methods of bulk measurement.Mainly can be divided three classes in the bulk measurement method: use continuous light intensity transmission measuring method; Use the frequency domain method of light intensity modulation; Use the time resolution method of light-pulse generator.Optical parametric detection method at body is comparatively complicated.Because the complicacy of human body often is subjected to the more multifactor influence in organizer inside in bulk measurement, causes the decline of measurement result accuracy.Stripped optical parameter measurement need be peeled away measuring samples and organizer, though on medical diagnosis, tend to bring certain misery to the patient, but with respect in the bulk measurement method, the method measuring accuracy height that exsomatizes and measure, the measuring method maturation is easy, therefore is used widely in organism optical research.

In the middle of present research, because the complex characteristics of organizer, people are seeking the method and the approach that can obtain organizer's optical parametric exactly.Time resolution diffuse reflection technology is with a branch of psec ultrashort pulse light incident biological tissue surface, concerns over time utilizing the high speed detector with picosecond resolution to record emergent light intensity with incidence point a distance.The tissue optical property parameters precision that this technology obtains is higher, but because signal to noise ratio (S/N ratio) is relevant with the photon number of measuring the interval, the short time interval and high signal to noise ratio (S/N ratio) are difficult to obtain simultaneously, and the measuring speed of the method is lower, and the instrument cost is expensive.Based on the spatial discrimination technology of continuous light the light in the organizer is distributed and to regard as and a kind of steady-state distribution that the time is irrelevant, in organizer's surface range by obtaining the photodistributed situation of change of diffuse reflection, to extrapolate the optical parametric of tissue.Coherence chromatographic imaging technology (OCT technology) utilizes the near infrared weak coherent light to shine tissue to be measured then based on principle of interference, interferes according to the relevant generation of light and surveys the tomography of tissue, thereby obtain the optical properties of tissue parameter.The OCT technology can realize the not damaged to biological tissue, real-time detection and imaging, and can be used for the structure analysis of vivo biological tissue and the measurement of characteristic parameter.But OCT technology penetration capacity for some organizer in application process is not strong, can only detect the pathology situation of shallow layer tissue's body.The integrating sphere technology is that application is more extensive, full-fledged in the existing detection technique, the accurate relatively optical parametric detection means of measurement result.The integrating sphere technology is the reflectivity by measuring samples, the measurement of diffuse transmittance and collimated transmittance, and then obtain the optical parametric of sample.The integrating sphere technology can be divided into simple integral playing skill art and double integrator playing skill art, for simple integral playing skill art, double integrator playing skill art can be eliminated the Effect of Environmental of bringing owing to the sample position conversion, can obtain the reflection and the intensity in transmission of sample under the condition that does not change sample position simultaneously.

How integrating sphere technology at present commonly used detects integrating sphere and combines with the anti-structure algorithm of optical parametric, and then anti-structure goes out the optical parametric of sample, and the anti-structure algorithm of optical parametric commonly used be against Monte Carlo algorithm (IMC) with against double method (IAD).Contrary Monte Carlo algorithm is the inverse process of Monte Carlo algorithm, is not subjected to the restriction of sample geometric configuration, boundary condition.But its calculated amount is huge, has limited the application of contrary Monte Carlo algorithm.The contrary algorithm that doubles has supposed that the setting up in the process of model the thickness of organizer is limited, and suppose in its each layer optical parametric all with distribution, sample surfaces infinitely extends.Its computing velocity has certain advantage than contrary monte carlo method, but under the situation of the albedo of sample less (absorption coefficient is bigger, and scattering coefficient is less), the error of calculation of contrary double method is bigger.The BP network is as a kind of multilayer feedforward neural network, and is simple in structure, is a kind of model that is most widely used in the neural network.The algorithm basic thought can be summarized as by two processes of backpropagation of the forward-propagating of signal and error to be formed, and during forward-propagating, the input sample imports into from input layer, passes to output layer after each hidden layer is successively handled.If the actual output and the desired output of output layer are not inconsistent, then change the back-propagation phase of error over to.APPLICATION OF MONTE CARLO METHOD is carried out the anti-structure of optical parametric in conjunction with neural network, because the good nonlinear fitting ability of neural network has been widened the optical parametric scope that anti-structure algorithm is suitable for.Simultaneously, the sample measurement value of the high emulation of employing Monte Carlo modelling and the mapping relations between the optical parametric are trained neural network, have also improved the precision of algorithm to the anti-structure of optical parametric of sample.This method has overcome the slow and restricted shortcoming of accommodation of speed of algorithms most in use, possesses good prospects for application.

Summary of the invention

The objective of the invention is to overcome the above-mentioned deficiency of prior art, a kind of tissue optical parameter measuring method is provided, comprise biproduct bulb separation measuring system and corresponding based on the anti-structure algorithm of the tissue optical parameter of neural network.Than other the method for measurement organizer optical parametric, among the present invention related system realized different wave length undertissue bulk optics parameter fast, measure in real time.

Technical scheme of the present invention is as follows:

A kind of quick multi-wavelength tissue optical parameter measuring device, comprise light-source system, have light path system, detection system, computing machine and the data acquisition system (DAS) of biproduct bulb separation, it is characterized in that, described light-source system comprises: four or more semiconductor laser can provide the LASER Light Source of four or more wavelength; Multiselect one photoswitch is used for described semiconductor laser is selected to switch; Signal generator is used for the selected laser that laser instrument produced is carried out high frequency modulated; Laser driver; Collimation lens is used for changing the diverging light that semiconductor laser sends into collimated light; Aperture diaphragm is used to limit the beam size of described collimated light; Described detection system comprises the detecting device that diffuses, diffuse transmission photodetector and three detecting devices of collimated light detecting device and back utmost point amplifier and lock-in amplifier, collimated light beam incides on the sample by the light inlet of reflection sphere, after the repeatedly reflection of the process that diffuses reflection sphere inwall, its light intensity signal is by the detector acquisition that diffuses on the reflection sphere; After light passed sample, diffuse transmission light was through the repeatedly reflection of transmission ball inwall, and its light intensity signal is obtained by the diffuse transmission photodetector on the transmission ball; The collimation transmitted light passes the light-emitting window of transmission ball, light intensity signal is after overdamping, by the collimated light detector acquisition by the collimated light detector acquisition, the signal of three detector acquisition is by post-amplifier and lock-in amplifier, is converted to digital signal and sends into computing machine and handle by data acquisition system (DAS).

The present invention provides a kind of trans-construction method that adopts above-mentioned measurement mechanism to realize simultaneously, comprises the following steps:

(1) sets up with monte carlo method and comprise three optical parametrics, be i.e. absorption coefficient, scattering coefficient and anisotropy factor and three measuring amount, the i.e. tranining database of the mapping relations between diffuse reflectance, diffuse transmittance and the collimated transmittance;

(2) according to measurement requirement, utilize multiselect one photoswitch, select semiconductor laser;

(3) preparation has the imitative body of standard of one group of tissue to be measured of fixed optics parameter,

(4) utilize described device, adopt the method for relative measurement, measure diffuse reflection spectrum, diffuse transmission spectrum and the collimation transmitted spectrum of the imitative body of described standard, obtain diffuse reflectance, diffuse transmittance and three measuring amount of collimated transmittance;

(5) utilize the measurement data of described standard being imitated body, the data in the described tranining database are proofreaied and correct, obtain new neural metwork training database.

(6) set up the BP neural network, input layer is three measuring amount of new neural metwork training lane database, and output layer is three corresponding three optical parametrics of measuring amount in this storehouse;

(7) utilize the data in the tranining database that the BP neural network is trained, meet the aimed at precision requirement up to the neural network result;

(8) adopt described device, obtain the measuring amount of tissue sample to be measured;

(9) neural network behind the application training, with the sample reflectivity that measures, diffuse transmittance and collimated transmittance, anti-structure goes out the optical parametric of sample.

Characteristics of the present invention are:

(1) the present invention has built the stripped optical parametric detection system of brand-new biproduct bulb separation.The extensibility of system is good, the precision height.

(2) in system, add photoswitch many each light sources are carried out switching controls.Settled four semiconductor lasers in the native system, adopted four to select a photoswitch to carry out the light source switching, and adopt the LabView program that the switching between the light source is controlled.The switch speed of photoswitch can reach 10ms, and the optical parametric that can satisfy sample carries out in real time, and quick needs of measuring have been widened the wavelength measurement scope of experimental system.

(3) system among the present invention adopts the sample cell of circulation sample introduction, can weaken in the fluid sample experimentation because the experimental error that the fluid sample precipitation is brought improves experimental precision.

(4) system of the present invention adopts the collection of LabView program realization data, and simultaneously to the photoswitch in the experimental system, lock-in amplifier, drivers etc. carry out software control, make things convenient for man-machine interaction.

(5) than existing optical parametric trans-construction method, adopt neural network algorithm to carry out the anti-structure of tissue optical parameter, improved computing velocity, widened optical parametric scope applicatory.

(6) use the BP neural network among the present invention and carry out optical parametric reconstruct,, significantly reduced because the influence of measuring error to bringing for the anti-structure of optical parametric because neural network itself possesses certain fault-tolerance.

(7) anti-structure algorithm is combined the variation of online reflection sample optical parametric that can be real-time with the software control of experimental system.

Related tissue optical parameter measuring system among the present invention, light source under the different wave length is added high_frequency sine wave to be modulated, and in the input part, adopt high-gain, broadband, low noise photoelectricity pre-amplification circuit, and by lock-in amplifier useful signal is extracted, direct current in the filtered signal and undesired signal, the signal to noise ratio (S/N ratio) of raising system.A plurality of instruments in the total system are controlled by LabView software, have strengthened the controllability of system.

Description of drawings

Fig. 1: biproduct bulb separation system chart.

Fig. 2: biproduct bulb separation systematic schematic diagram.

Fig. 3: biproduct bulb separation system source system chart.

Fig. 4: BP network structure.

Fig. 5: neural network procedure process flow diagram.

Embodiment

When related system carries out the tissue optical parameter measurement among the present invention, at first measure the reflectivity of sample by integration sphere light source system, diffuse transmittance and collimated transmittance use neural network algorithm to carry out modeling and data fitting then, obtain the optical parametric of sample.

Measuring system related among the present invention is mainly formed (see figure 1) by following several parts:

(1) signal generator 8 that the light-source system 1 of collimated light source is provided and is used for light source is modulated for experimental system.

(2) the biproduct bulb separation measuring system 2 that the reflectivity and the transmissivity of sample detected.

(3) light signal in will testing is converted to the photoelectric detector system 3 of electric signal.

(4) back utmost point amplifier 4 and the lock-in amplifier 5 that ultra-weak electronic signal is amplified.

(5) total system is carried out the computing machine 7 of software control and data presentation and processing, and system power supply 6.

Among the present invention light-source system has been added four and selected a photoswitch, can in system, place a plurality of semiconductor lasers, and, satisfied real-time measurement requirement, and widened the sensing range of system sample by the quick conversion of LabView software control light source.It is specifically formed as Fig. 3, comprises semiconductor laser, driver, and four select a photoswitch, attenuator, aperture diaphragm, several sections such as signal generator:

(1) semiconductor laser 9, semiconductor laser 10, semiconductor laser 11, semiconductor laser 12: the semiconductor laser light resource that is provided with a plurality of wavelength in the system.Optical source wavelength can be selected in 633nm, 753nm, 808nm, 904nm, 1060nm, 1150nm, 1310nm, 1521nm.Widened the measurement range of system.

(2) signal generator 8: can produce the light modulated of different frequency to the different wavelength of laser light source, the light source after the use high frequency modulated carries out the signal to noise ratio (S/N ratio) that optical detection helps to improve system.

(3) laser driver 13: for semiconductor laser provides the optimum current that is fit to its work and controls its working temperature, guarantee the stable of laser works.

(4) four select a photoswitch 14: can change fast light source, the fastest switching rate can reach several milliseconds, satisfy system to the optical parametric of sample carry out in real time, the requirement of fast detecting.

(5) collimation lens 15: the diverging light that semiconductor laser can be sent changes collimated light into, reduces the angle of divergence of light source.

(6) aperture diaphragm 16: the aperture of confine optical beam.

(7) attenuator 17: because the collimation transmitted light intensity is bigger, attenuator is used to the light intensity that decays, and prevents that photoelectric detector 3 is saturated.

Integration sphere light source system (as Fig. 2) and thereafter utmost point amplification system form by following each several part:

(1) biproduct bulb separation: comprise integrating sphere 1 (18) and integrating sphere 2 (19) two parts, its inwall has highly-reflective coating, can be in the uniform light intensity of the inner formation of spheroid.

(2) photodetector system 3: comprise photoelectric detector 1 (20), to the reflected light of sample, transmitted light and collimation transmission light intensity detect, and light signal is converted to electric signal respectively for photoelectric detector 2 (21), photoelectric detector 3 (25).

(3) lock-in amplifier 22: signal can be extracted from noise, possess the ability of improving signal to noise ratio (S/N ratio) preferably, can eliminate the influence of bias light.

Included other instruments in this system:

(1) computing machine 7: be used for data acquisition, and data are carried out analyzing and processing, realize tissue optical parameter reconstruct on the software.

(2) sample cell 23: form sandwich shape formula sample cell by two glass sheet and sample, be used to hold sample, and have multiple thickness can supply experimental selection.

(3) peristaltic pump 24: be used for the sample introduction of liquid, make laboratory sample be in recurrent state, can prevent the error that the fluid sample precipitation is brought.

(4) D.C. regulated power supply 6: for photoelectric detector provides stabilized voltage supply.

The present invention with the double integrator ball measurement method with combine based on the optical parametric reconstruct of BP neural network.According to the simulation of light transmission model in the tissue or have the measurement of the imitative body of known fixed tissue optical parameter (absorption coefficient, scattering absorb, anisotropy factor), can obtain each optical parametric (μ of organizer _a, μ _s, g) and reflectivity, the forward model between diffuse transmittance and the collimated transmittance.According to correlation theory, the propagation relative complex of light in organizer, therefore when carrying out the anti-structure of optical parametric, measured value (R _d, T _d, T _c) and optical parametric (μ _a, μ _s, g) relation between is non-parsing, is difficult to obtain result accurately by fitting of a polynomial or other match difference approach.Neural network has very strong non-linear mapping capability, and the network hidden layer and the transport function of neural network reasonably designed, and just can satisfy the needs of optical modeling and data fitting.

Among the present invention, adopt the method for Monte Carlo simulation to set up mapping relations between the optical characteristics of the sample that measures in optical parametric and the reality.The basic framework of Monte Carlo calculations model mainly comprises the determining of photon arbitrary width, photonic absorption and the description of scattering process and the descriptive statistics of measured physical quantity.The specific implementation of relevant this method, can vide infra: MCML-Monte Carlo modeling of light transport in multi-layered tissues/Lihong Wang, Steven L.Jacques, Liqiong Zheng //Computer Methods and Programsin Biomedicine, Volume 47, Issue 2, and July 1995, Pages 131-146

The transmission course of Monte Carlo method simulated light in tissue just writes down the trace of each photon in tissue in a large number, and simulates the trace of each photon in biological tissue, and its step divided for four steps on the whole:

(1) determines initial trace point according to the incident condition; Determine the step-length of photon motion;

(2) determine direction that photon is advanced and the position of next time colliding;

(3) determine absorption and scattered portion at this position photon;

(4) returned for the 3rd step.

So cycle calculations less than a certain setting value, just finishes the tracking to this photon up to the photon weight when perhaps photon overflows biological tissue's upper and lower surface.Return the first step then and write down another photon, finish up to whole tracking of the photon number that sets.Adopt the MC model to carry out that relation has advantage between tissue optical parameter and the measuring amount: (1) realizes simple, simulation accurately, its operational process only needs several indicative of local optical parameters; (2) can simulate the light dissemination under respectively of random geometry, boundary condition and optical parametric; (3) can be in output quantity the poisson noise performance of introducing intrinsic of nature, thereby effectively set up the test statistics model.

Transport direct, the effective and the most believable method of actual physics process as a simulated photons, Monte Carlo simulation is playing crucial effect aspect the validation verification of the spectral detection of using the light transmission models based on other and imaging technique scheme.Practice simultaneously is proof also, determines that with Monte Carlo method the transmission of photon in biological tissue is the most accurate, and is best with the degree of agreement of actual measurement.

The present invention utilizes the nonlinear fitting and the predictive ability of BP neural network, trains by adopting mass data set pair network, and the network that the input of actual sample measuring amount is trained is the measurable tissue optical parameter that meets with a response then.The BP neural network structure adopts an input layer, two hidden layers, and an output layer (network structure is seen Fig. 4) carries out to network that input layer is provided with 600 nodes, input vector P in the training process _k=(a ₁, a ₂... a _n), the measured value of representing laboratory sample is the training input set of standard.Output layer has 600 nodes.The output vector C of network _k=(y ₁, y ₂..., y ₆₀₀) to represent the tissue optical parameter of laboratory sample be the training output collection of standard.In the system algorithm design, consider the systematic survey parameter characteristic---measurement of reflectivity, diffuse transmittance and collimated transmittance (R _d, T _d, T _c) scope is in (0,1) interval, and measuring amount and optical parametric be matrix form, design this BP network and contain two hidden layers.Theoretical confirmation, the network that contains two hidden layers can approach arbitrary function.Simultaneously,,, when adopting bigger modeling sample collection, need the number of hidden nodes response ratio bigger, but also will avoid generation simultaneously owing to the too much over-fitting phenomenon that causes of latent node number for improving the network precision according to the designing requirement of BP neural network.At 600 training sets, experiment showed, that the network mapping ability is the strongest, and can avoid the over-fitting phenomenon when 25 latent nodes of design.Concrete network structure is seen Fig. 4.

Neural network feedforward system among the present invention, set up according to the following step:

The first step: adopt Monte Carlo method to set up the sample data storehouse of standard.Set up optical parametric (μ _a, μ _s, g) with measuring amount (R _d, T _d, T _c) between mapping relations.And the data in this database are used for carrying out the training of neural network.

Second step: set up the BP neural network that contains two hidden layers.The input signal of this BP neural network is (R _d, T _d, T _c), represent reflectivity, transmissivity and the collimated transmittance of organizer or sample.Output signal is the tissue optical parameter (μ of sample _a, μ _s, g), absorption coefficient, scattering coefficient and the anisotropy factor of difference representative sample or organizer.

Approach effect preferably in order to have, the hidden layer transport function in the neural network adopts " tansig " and " purelin " function respectively." tansig " function is a tanh S type transport function, and its form of calculation is:

n＝2/(1+exp(-2n))-1

Be fit to Nonlinear Modeling and prediction requirement, " purelin " function is a linear transfer function.Learning function adopts " learngdm ", is gradient decline momentum learning function, adopts " traingd " gradient descent algorithm in the network training, and performance function " mse " comes the evaluating network performance with square error.

The 3rd step: the standard database that adopts monte carlo method to set up is the training signal of neural network error signal.Set up (R _d, T _d, T _c) → (μ _a, μ _s, mapping relations g).In the training process of neural network, with input sample P _k=(a ₁ ^k, a ₂ ^k... a _n ^k) connection weight w _IjWith threshold value θ _jCalculate the input s of each unit, middle layer _j, use s then _jCalculate the output b of each unit, middle layer by transport function _j

$s_j=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{a}_i-{\mathrm{\&theta;}}_j$ j＝1，2，…，25

b _j＝f(s _j) j＝1，2，…，25

Utilize the output b in middle layer _j, connection weight v _JtAnd threshold gamma _tCalculate the output L of each unit of output layer _t, utilize transport function to calculate the corresponding C of each unit of output layer then _t

L _t＝∑v _jtb _j-γ _t t＝1，2，…，600

C _t＝f(L _t) t＝1，2，…，600

Utilize the network objectives vector T _k=(y ₁ ^k, y ₂ ^k..., y ₆₀₀ ^k), the actual output C of network _t, each unit vague generalization error d of calculating output layer _t ^k

$d_t^k=(y_t^k-C_t)\&CenterDot;C_t(1-C_t)$ t＝1，2，…，600

Utilize connection weight v _Jt, the vague generalization error d of output layer _t ^kOutput b with the middle layer _jCalculate the vague generalization error e of each unit, middle layer _j ^k

$e_j^k=[\underset{t=1}{\overset{600}{\mathrm{\&Sigma;}}}{d}_t\&CenterDot;v_{\mathrm{jt}}]b_j(1-b_j)$

Utilize the vague generalization error d of each unit of output layer _t ^kOutput b with each unit, middle layer _jRevise connection weight v _JtAnd threshold gamma _t

$v_{\mathrm{jt}}(N+1)=v_{\mathrm{jt}}\left(N\right)+\mathrm{\&alpha;}\&CenterDot;d_t^k\&CenterDot;b_j$

${\mathrm{\&gamma;}}_t(N+1)=v_{\mathrm{jt}}\left(N\right)+\mathrm{\&alpha;}\&CenterDot;d_t^k$

t＝1，2，…，600，j＝1，2，…，25，0＜α＜1

Utilize the vague generalization error e of each unit, middle layer _j ^k, the input P of an input layer unit _k=(a ₁, a ₂... a _n) revise connection weight w _IjWith threshold value θ _j

$w_{\mathrm{ij}}(N+1)=w_{\mathrm{ij}}\left(N\right)+\mathrm{\&beta;}\&CenterDot;e_j^k\&CenterDot;a_i^k$

${\mathrm{\&theta;}}_j(N+1)={\mathrm{\&theta;}}_j\left(N\right)+\mathrm{\&beta;}\&CenterDot;e_j^k$

i＝1，2，…，n，j＝1，2，…，25，0＜β＜1

The 4th step: adopt integration sphere light source system to obtaining the measured value (R of sample _d, T _d, T _c).

The 5th step: adopt above-mentioned BP neural network that the optical parametric of sample is reconstructed, finish (R _d, T _d, T _c) → (μ _a, μ _s, calculating g).

The scope of application of optical parametric reconstruct has been widened in the application of neural network aspect optical parametric reconstruct.Simultaneously, neural net method has overcome also that computing time of optical parametric trans-construction method in the past is long, advantage such as defectives such as the range limited system of optical parametric reconstruct have wide accommodation, and are precise and stable, flexible.

The method of relative measurement is adopted in the measurement of sample reflectivity of the present invention and transmissivity, the transmitted light intensity of the reflective light intensity of measurement standard reflecting plate and air at first, the reflective light intensity of measuring samples and transmitted light intensity are deducted reflectivity and transmissivity that noise factor calculates sample then.Bring the optical parametric that reconstructing method draws sample then into.

System of the present invention can realize that measurement light source switches fast, and optical parametric detects in real time under the different wave length.The Lights section adopts the semiconductor laser and the multi channel selecting photoswitch of a plurality of different wave lengths, expands a plurality of measurement wavelength, and photoswitch uses software control, and can satisfy the needs of real-time detection switching time between two wavelength in 10ms.Be to improve measuring accuracy, the incident light in the system should satisfy unlimited thin directional light condition, adopts a self-focusing optical fiber beam path alignment device and small-bore diaphragm that light beam is collimated in the invention and limits, to satisfy measurement requirement.After light incided sample, reflected light and transmitted light were collected by two integrating spheres respectively, and after the inner process of integrating sphere highly-reflective coating repeatedly reflected, the light intensity in the integrating sphere evenly distributed.By the preposition Photoelectric Detection amplifying circuit of low noise high-gain that is placed on the integrating sphere inwall light intensity signal is converted into electric signal,, and behind the data collector result is imported computer, show and do further processing through the phase-locked amplifying circuit of level later.In carrying out optical signal detecting and amplification process, adopted ac modulation light signal and phase lock amplifying technology, can effectively suppress the influence that various noises in the circuit and power frequency are disturbed, improve the signal to noise ratio (S/N ratio) that feeble signal is measured.According to calculating, use the phase-locked signal to noise ratio (S/N ratio) of measuring the back system to improve more than 20 times.The anti-structure partial data of optical parameter is handled and is comprised and will measure the reflectivity that original vol is converted to sample, and diffuse transmittance and collimated transmittance are imported the optical parametric that trained BP neural network carries out obtaining after the data fitting reconstruct sample: absorption coefficient μ then _s, absorption coefficient μ _a, anisotropy factor g etc.With restructing algorithm embedded system Control Software part, can handle measured value automatically, and the optical parametric Computer Storage be shown the variation of the optical parametric of online reflection sample.Concrete experimental procedure is divided following components:

Measure portion, measuring principle is seen Fig. 2:

(1) pre-thermal light source is adjusted light path and is reached collimation.

(2) use standard reflecting plate with integrating sphere 1 (18) and integrating sphere 2 (19) go out, light inlet blocks, and measures the interior dark noise of two integrating spheres, is designated as Rd-1, Tt-1;

(3) remove standard reflecting plate, light is measured the signal in the integrating sphere by two integrating spheres.Signal and dark noise are on close level under the situation of adjustment beam path alignment;

(4) standard reflecting plate (80%) is placed the light-emitting window of integrating sphere 1 (18), the measuring light signal is as the reflection measurement reference value, and note is Rd-f; Open integrating sphere 1 light-emitting window, will reflect the light-emitting window that lid places integrating sphere 2 (19) transmission balls, the measuring light signal is Td-f as diffuse transmission witness mark value note; Open that two integrating spheres go out, light inlet, adjust the collimation detector location and make and collimation transmission measurement signal maximum obtain collimating light intensity reference value Tc-f;

(5) in sample cell, place the sample that to measure.Open integrating sphere 2 light-emitting windows, three detecting devices write down the reflection strength value of sample respectively, the diffuse transmission intensity level, and collimation intensity in transmission value is designated as Rd-s, Td-s, Tc-s.

(6) cover transmission ball light-emitting window.Detecting device 1 obtains the intensity of reflected light size Rd-1-s of sample, and detecting device 2 obtains the transmitted intensity size Tt-s of sample;

Experimentation obtains the noise Rd-1 of system, Tt-1; Obtain the reference light intensity Rd-f of system, Td-f, Tc-f obtains measured value Rd-s, Rd-1-s, Td-s, Tc-s, Tt-s behind the placement sample

Then obtain the reflectivity of sample $V_1\%=\frac{R_{d-1-s}-R_{d-1}}{(R_{d-f}-R_{d-1})/80\%}$

The transmissivity of sample $V_2\%=\frac{T_{t-s}-T_{d-1}}{T_{d-f}-T_{d-1}}$

The collimated transmittance of sample $V_3\%=\frac{T_{c-s}-T_{c-1}}{T_{t-f}-T_{c-1}}$

Under the lower situation of noise level, above formula can be reduced to:

The reflectivity of sample $V_1\%=\frac{R_{d-1-s}}{R_{d-f}/80\%}$

The transmissivity of sample $V_2\%=\frac{T_{t-s}}{T_{d-f}}$

The collimated transmittance of sample $V_3\%=\frac{T_{c-s}}{T_{c-f}}$

Algorithm modeling and match part, concrete neural network algorithm flow process is seen Fig. 5:

(1) optical parametric simulated data collection (μ is set _a, μ _s, g), utilize the Monte Carlo simulation program, obtain reflectivity, transmissivity and the collimated transmittance of the sample that this simulated data set pair answers, utilize neural networks principles to set up (μ _a, μ _s, g) → (R _d, T _d, T _c) Nonlinear Mapping relation

(2) preparation has fixed optics parameter (μ _a, μ _s, one group of standard g) is imitated body, utilizes the integrating sphere measuring system to obtain its reflectivity and diffuse transmittance and collimated transmittance, and itself and analog quantity are compared, and the simulated data collection is proofreaied and correct, and obtains new neural metwork training collection.

(3) make up initialization BP neural network, utilize training set that network is trained, less than set threshold value, the BP neural network after the training is used for data fitting and the anti-structure of optical parametric up to error of fitting.

(4) with the sample reflectivity and the transmissivity neural network that input trains as input quantity that measure, just can obtain corresponding sample tissue optical parametric (μ by anti-structure _a, μ _s, g).

Software control part

Among the present invention the integration and the controllability of enhanced system, designed Control Software based on LabView, by GPIB between computing machine and the various instrument or RS232 communication interface, various devices in the various systems to be controlled, Design of Man-Machine Interaction Interface is operated system more easily.The software several main parts comprises:

(1) control section: according to the gpib interface communications protocol of computing machine and laser driver, realize the control of computing machine, comprise the selection of different passages, the control of drive current and semiconductor laser working temperature etc. to laser driver; Between computing machine and the photoswitch, by the RS232 interface, realize the control to photoswitch, can switch different photoswitch passages fast, realize the quick selection to wavelength, the switch speed between two wavelength is less than 10ms.

(2) part of data acquisition: carry out communication by gpib interface between computing machine and the digital phase-locking amplifier, program can be controlled the locking frequency of lock-in amplifier on the one hand, integral time etc., lock-in amplifier enters computing machine with digital data transmission on the other hand, shows and does subsequent treatment

Anti-structure algorithm interface: the anti-structure algorithm that will design is embedded in the control master routine, and measured value is imported the organism optical that algorithm process obtains measuring samples after treatment, and shows in real time, is stored in the file simultaneously.

Claims (3)

1. quick multi-wavelength tissue optical parameter measuring device, comprise light-source system, have light path system, detection system, computing machine and the data acquisition system (DAS) of biproduct bulb separation, it is characterized in that, described light-source system comprises: four or more semiconductor laser can provide the LASER Light Source of four or more wavelength; Multiselect one photoswitch is used for described semiconductor laser is selected to switch; Signal generator is used for the selected laser that laser instrument produced is carried out high frequency modulated; Laser driver; Collimation lens is used for changing the diverging light that semiconductor laser sends into collimated light; Aperture diaphragm is used to limit the beam size of described collimated light; Described detection system comprises the detecting device that diffuses, diffuse transmission photodetector and three detecting devices of collimated light detecting device and back utmost point amplifier and lock-in amplifier, collimated light beam incides on the sample by the light inlet of reflection sphere, after the repeatedly reflection of the process that diffuses reflection sphere inwall, its light intensity signal is by the detector acquisition that diffuses on the reflection sphere; After light passed sample, diffuse transmission light was through the repeatedly reflection of transmission ball inwall, and its light intensity signal is obtained by the diffuse transmission photodetector on the transmission ball; The collimation transmitted light passes the light-emitting window of transmission ball, light intensity signal is after overdamping, by the collimated light detector acquisition by the collimated light detector acquisition, the signal of three detector acquisition is by back utmost point amplifier and lock-in amplifier, is converted to digital signal and sends into computing machine and handle by data acquisition system (DAS).

2. a trans-construction method that adopts the described measurement mechanism of claim 1 to realize is characterized in that, comprises the following steps:

(1) sets up the tranining database comprise the mapping relations between three optical parametrics and three measuring amount with monte carlo method, described three optical parametrics are absorption coefficient, scattering coefficient and anisotropy factor, described three measuring amount are diffuse reflectance, diffuse transmittance and collimated transmittance;

(2) according to measurement requirement, utilize multiselect one photoswitch, select semiconductor laser;

(3) preparation has the imitative body of standard of one group of tissue to be measured of fixed optics parameter,

(4) utilize described device, adopt the method for relative measurement, measure diffuse reflection spectrum, diffuse transmission spectrum and the collimation transmitted spectrum of the imitative body of described standard, obtain diffuse reflectance, diffuse transmittance and three measuring amount of collimated transmittance;

(5) utilize the measurement data of described standard being imitated body, the data in the described tranining database are proofreaied and correct, obtain new neural metwork training database.

(6) set up the BP neural network, input layer is three measuring amount of new neural metwork training lane database, and output layer is three corresponding three optical parametrics of measuring amount in this storehouse;

(7) utilize the data in the tranining database that the BP neural network is trained, meet the aimed at precision requirement up to the neural network result;

(8) adopt described device, obtain the measuring amount of tissue sample to be measured;

(9) neural network behind the application training, with the sample reflectivity that measures, diffuse transmittance and collimated transmittance, anti-structure goes out the optical parametric of sample.

3. trans-construction method according to claim 2, it is characterized in that set up the BP neural network that contains two hidden layers, the number of hidden nodes is 25, the hidden layer transport function adopts " tansig " and " purelin " function respectively, adopts gradient decline weights learning function BP network.

Images