Human body system modeling and evaluation method based on MCMD ARMA model

Abstract

The invention discloses a human body system modeling and evaluation method based on an MCMD ARMA model, belongs to the technical field of system medicine, signals and system application, and particularly relates to a human body system model which is achieved based on the MCMD ARMA model and takes a tongue image as system output and a model evaluation method. According to the method, the problems that how to further apply the MCMD ARMA model to modeling of a human body system to obtain mathematical representation of a human body health state and how to evaluate the accuracy of the model are solved. The human body system is described through the MCMD ARMA model, the accuracy of the model is evaluated by comparing the difference between the reconstructed signal and the measurement signal, and the effectiveness of the human body state is revealed through the classification experiment verification model.

Description

Technical Field

[0001] The present application relates to the field of system medicine, signal and system application technology, and specifically to a modeling and model evaluation method for a human body system based on a MCMD ARMA model. Background Art

[0002] TCM believes that people live in a natural environment and are constantly affected by the external environment. When the impact of the external environment exceeds the threshold of human function adaptation, dysfunction and disease will occur. When looking at diseases, patients and the natural environment should be examined together. TCM regards people as a whole, the human body as a system affected by the environment, and looks at diseases from the perspective of the correspondence between people and heaven and earth. This is the principle of "examining the inside and outside" in diagnostics. TCM believes that clinical data contains a wealth of information related to the human system, and regards the clinical information obtained through the four examinations of observation, auscultation, inquiry, and palpation as the output of the human system, which is the response of the human body to internal and external influences. TCM comprehensively judges the health status of people through the four examinations and completes the identification of the human system. The idea of ​​system medicine is similar to TCM. System medicine believes that mathematical models can be used to fit the human system, mathematical expressions can be used to represent the human body, and clinical data can be used as the output of the human system. The human state can be judged through clinical information. Many studies have applied the idea of ​​system medicine.

[0003] In a broad sense, a system is a signal mapping processor that receives input signals and responds to output signals according to certain rules. For the human body system, there are many factors that affect the human body system. External environmental factors (such as cold and heat syndromes, etc.) and internal factors (such as emotions, stress, etc.) may cause changes in human health. But generally speaking, only extreme emotional changes or long-term emotional accumulation will cause changes in human state. In most cases, disease is the response of the human body to external environmental factors. Various clinical data show that the response of the human body to changes in the external environment within a certain period of time can be approximated as a linear time-invariant system (Linear Time-Invariant system, LTI system). For example, a person's heartbeat pulse exhibits time-invariant characteristics for a short period of time when resting, while a person's pulse changes linearly with the intensity of exercise during exercise. Chinese medicine tongue diagnosis evaluates the health status of the human body by observing the tongue. The clinical information of the tongue image is the response of the human body system to external input. This application models the clinical information of the tongue image as the output of the human body system. The text considers the tongue image at a certain time node, which does not involve the time dimension. The LTI model is sufficient for basic behavior prediction and analysis. Of course, more complex system designs can analyze the human body from a more comprehensive perspective. The LTI model may not accurately capture more complex biological dynamics and pathological changes, and simplifies the complexity of the human system.

[0004] This application is based on the ideas of systems medicine and traditional Chinese medicine, and simplifies the modeling of the human body system into an LTI system. From the perspective of traditional Chinese medicine tongue diagnosis, the clinical data of tongue diagnosis is used as the output of the human body system. Taking into account the impact of the environment on the human body, the environmental impact is a random factor. In order to simplify the analysis, this application models the external random environmental factors as Gaussian white noise as the input of the human body system. The LTI system can be represented by a linear equation, and when the input and output signals of the LTI system are wide-sense stationary signals, the system can be characterized by an auto-regressive moving average (ARMA) model. A meaningful tongue image signal should be a wide-sense stationary signal. On the one hand, the ARMA model has been used to characterize human body signals such as heartbeat and pulse, and experiments have proved the feasibility of the ARMA model for human body signal characterization. On the other hand, the ARMA model has been applied to the modeling of image signals such as faces. According to experimental results, the ARMA model can well characterize facial images, and the extracted features can be used for face recognition. Summary of the invention

[0005] The technical problem to be solved by this application is: how to implement ARMA modeling based on high-dimensional images, and how to further apply the MCMD ARMA model to the modeling of the human body system to obtain a mathematical representation of the human health state and evaluate the accuracy of the model. This application provides a method for modeling, solving and evaluating a human body system based on the MCMD ARMA model to solve the above problems. This application obtains a mathematical description of a human body system based on the MCMD ARMA model, estimates the model parameters using the least squares method as a representation of the human health state, and performs performance evaluation.

[0006] The ARMA model is a classic method for LTI system identification. This application models the human body LTI system based on the ARMA model. Inspired by the ARMA model and based on the idea of ​​systems medicine, this application proposes a multi-channel multi-dimensional autoregressive moving average (MCMD ARMA) model. From the perspective of traditional Chinese medicine tongue diagnosis, this application realizes human body system modeling with tongue images as system output. This application is implemented through the following technical solutions: This application provides a human body system modeling method based on the MCMD ARMA model, and the method includes:

[0007] Step 1: Select the subjects, determine the type of clinical data to be collected and place the corresponding data collection equipment;

[0008] Step 2: Acquire the collected sample information received by the data acquisition device, wherein the collected sample information is the relevant clinical medical information of the subject;

[0009] Step 3: Input sample information into the MCMD ARMA model and transform it into multiple single-channel multidimensional ARMA models. The model order is determined by the Akaike Information Criterion (AIC) and the unknown model parameters are estimated by the least squares method.

[0010] Step 4: Establish a multidimensional ARMA model for each single-channel image, determine the model order by the Akaike Information Criterion (AIC), estimate the unknown model parameters by the least squares method, and obtain the ARMA model corresponding to the single-channel clinical data;

[0011] Step 5: Use the estimated MCMD ARMA model to reconstruct the measured clinical medical signal, and compare the difference between the reconstructed signal and the measured signal to evaluate the accuracy of the model.

[0012] This application has the following advantages and beneficial effects:

[0013] Through the above steps, the present application realizes the characterization of the human body system through the MCMD ARMA model, and evaluates the accuracy of the model by comparing the difference between the reconstructed signal and the measured signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of image-based 2D AR (3, 2) modeling at point x (n, m) provided in an embodiment of the present application.

[0015] Figure 2 Schematic diagram of R, G, B single channel images of the RGB tongue image provided in an embodiment of the present application.

[0016] Figure 3 A comparison diagram of the real measurement signal of the tongue image provided in an embodiment of the present application (top) and the reconstructed signal of the tongue image obtained after modeling the selected experimental object based on the MCMD ARMA model (bottom).

[0017] Figure 4 This is the NRMSE mean of the reconstructed signal and the measured signal of the three channels of 230 tongue images provided in the embodiment of the present application.

[0018] Figure 5 Schematic diagram of the clinical tongue image acquisition device provided in an embodiment of the present application.

[0019] Figure 6 A schematic diagram of the tongue image segmentation effect based on U-Net provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Before any embodiment of the present application is described in detail, it should be understood that the application of the present application is not limited to the details of the structures shown in the following description or drawings. The present application may adopt other embodiments and may be implemented or executed in various ways. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative improvements are within the scope of protection of the present application.

[0021] Tongue diagnosis is an important basis for TCM diagnosis and one of the most widely used methods in TCM diagnosis. Tongue diagnosis is to judge the patient's health status by observing the patient's tongue color, tongue texture, tongue coating, tongue moisture, tongue tooth marks, etc., and to diagnose and evaluate the disease. According to the clinical experience of TCM, doctors can understand the patient's condition based on the color and texture of the tongue, so as to prescribe the right medicine and treat the disease. Due to its non-invasiveness, convenience and intuitiveness, tongue diagnosis is one of the important diagnostic methods in TCM and has high research value.

[0022] According to various clinical data, within a limited observation time, the human body's response to changes in the external environment is approximately linear and time-invariant, so a linear time-invariant system (LTI) can be used to approximately describe the reactions and changes in the human body system. In addition, the characterization of the human body system based on mathematical models is of great significance for further detailed parameter analysis of the human body system. The ARMA model has relevant applications for characterizing human body signals and image signals, and its feasibility has been proven.

[0023] The method adopted in this application is to propose an image-based MCMD ARMA model, and based on the MCMD ARMA model, human body system modeling with tongue images as system output is realized.

[0024] In the modeling of the human system based on the MCMD ARMA model, tongue diagnosis in traditional Chinese medicine is viewed from the perspective of system medicine. Gaussian white noise, an external random factor on the human body, is modeled as the input of the human system, and clinical data of tongue images are modeled as the output of the human system. Tongue diagnosis in traditional Chinese medicine uses clinically observed tongue data to judge the health status of the human body, and the human body is the system that needs to be identified. The steps are as follows:

[0025] Step 1: Collect clinical data and place corresponding data collection equipment;

[0026] Step 2: Obtain the sample information received by the data acquisition device. The sample information is the relevant clinical medical information of the subject. The data of each subject is multi-dimensional and multi-channel data. Where L 1 ,…,L M express The size in each dimension, K represents the number of channels. For example, an RGB image with a length and width of 20×30 can be represented as

[0027] Step 3: Modeling MCMD ARMA model based on multi-dimensional and multi-channel data;

[0028] For M-dimensional K-channel tensor data Divide the model into equal-sized blocks according to its AR order As mentioned above, I 1 ,…,I M express The size in each dimension, K represents the number of channels, x′(i 1 ,…,i M ,k) represents a tensor At point (i 1 ,i 2 …,i M ,k) element value; AR represents the autoregressive model.

[0029] Similarly, for the original data Consider noise tensors of equal size Divide the MA into noise tensor blocks according to its order Among them J 1 ,…,J N express The size in each dimension, K represents the channel, and MA represents the self-sliding average model.

[0030] As system input, consider an M-dimensional K-channel ARMA system, defined as follows:

[0031]

[0032] in, and are the coefficient tensors of the AR and MA parts respectively; Represents the AR parameter tensor and data tensor The tensor contraction product of the first M dimensions of is a second-order tensor, i.e., a K×K matrix, with matrix element index express Tensor parameters for channels and The data tensor of the channel The tensor contraction product of can be expressed as The formula is as follows:

[0033]

[0034] Similarly, Represents the MA parameter tensor and the noise tensor The tensor contraction product of the first N dimensions is also a K×K matrix;

[0035] For the convenience of writing, define the AR part of the model as The MA part is The expression is written as:

[0036]

[0037] in Symbols indicate definitions.

[0038] The model coefficients of different individuals must be different, and the model coefficients of the same person in different periods and different physical conditions are also different. Therefore, the MCMD ARMA model can model the human body system based on personalized parameters, and can provide a personalized auxiliary evaluation standard for doctors to diagnose the health status of patients.

[0039] The present invention provides a method for estimating individualized parameters of a MCMD ARMA model.

[0040] Step 1: Collect clinical data of tongue images and place corresponding data acquisition equipment;

[0041] Step 2: Obtain the sample information received by the data acquisition device. The sample information is the clinical information of the tongue image of the subject. The tongue image data of each subject is two-dimensional three-channel data.

[0042] Step 3: Human body system model based on MCMD ARMA with tongue image as output;

[0043] Tongue image data At the reference points x(n,m,1), x(n,m,2) and x(n,m,3), use (p 1 +1)×(p 2 +1)×3 size sliding window takes values ​​in 1 / 4 support area of ​​the image, and the image data block can be obtained. The same is true for noisy data According to (q 1 +1)×(q 2 +1)×3 size sliding window takes values ​​in the 1 / 4 support area, and the noise tensor data block can be obtained where max{p 1 ,q 1}<n≤L 1 -1,max{p 2 ,q2}<m≤L 2 -1. Tongue image data block and noise block The AR and MA part equations are as follows

[0044]

[0045] where p 1 ,p 2 ,q 1 ,q 2 is the two-dimensional ARMA order, which is determined by traversal in the experiment.

[0046] Step 4: Convert the MCMD ARMA model into multiple single-channel multidimensional ARMA models;

[0047] Slice the data according to different channels and write it in matrix form

[0048]

[0049] Among them A i ,B j The expression for (1≤i,j≤3) is as follows

[0050]

[0051] Step 5: Design optimization function solution model;

[0052] Tongue image data is two-dimensional three-channel data, then and The nine components correspond to nine two-dimensional ARMA systems, namely 2D ARMA systems, and the model residuals The expression is

[0053]

[0054] Since the three channels of R, G, and B are orthogonal, there is no need to consider the coherence between the channels. The optimization expression is min(tr(ε)), where tr(·) is the trace of the matrix, that is, tr(ε) = e 11 +e 22 +e 33 The original model can be converted into the following three 2D single-channel models, namely three 2D ARMA models for solution

[0055] 1 ,X 1 >= 1 ,W 1 >

[0056] 2 ,X 2 >= <B​​​2 ,W 2 >

[0057] 3 ,X 3 >= 3 ,W 3 >

[0058] Step 6: Use Akaike information to accurately determine the model order AIC;

[0059] AIC=-2*ln(L)+2*k

[0060] Among them, L is the likelihood function value of the maximum likelihood estimate of the model, and k is the number of parameters to be estimated;

[0061] Step 7: For each single-channel image of the tongue image, approximate the 2D ARMA model with a 2D AR model;

[0062] The image-based ARMA model is defined to model the 1 / 4 area of ​​the image. The image-based 2D AR model is as follows: Figure 1 Therefore, for a single-channel two-dimensional image data X = {x(i 1 ,i 2 ):0≤i 1 ≤L 1 -1,0≤i 2 ≤L 2 -1}, the ARMA model equation established at the image point x(n,m) can be written as

[0063] x(n,m)+φ (n,m) θ=w(n,m)

[0064] in θ is the model parameter vector to be solved, p 1 ,p 2 ,q 1 ,q 2 is the two-dimensional ARMA order. Since in the image X 1 For n=P+1,...,L 1 -1; m=Q+1,...,L 2 -1,P>max(p 1 ,q 1 ),Q>max(p 2 ,q 2 ) pixel point x(n,m) equations are all valid, and it is expanded into a matrix form to obtain

[0065] x+Φθ=w

[0066] ​​Where x is the output signal vector, θ is the model parameter vector to be solved, and w is the input signal vector. The expression is:

[0067] x=[x(P+1,Q+1),x(P+2,Q+1)…,x(L 1 -1,L 2 -1)] T

[0068] w=[w(P+1,Q+1),w(P+2,Q+1)…,w(L 1 -1,L 2 -1)] T

[0069]

[0070] In order to solve the model input term {w(n,m)}, AR(K 1 ,K 2 ) model approximates the ARMA model, and the AR(K 1 ,K 2 ) The model expression is

[0071]

[0072] Similarly, for n = K 1 ,...,L 1 -1; m = K 2 ,...,L 2 -1 condition of the original image pixel x(n,m) equation is established, and it is expanded into a matrix form to obtain

[0073] x+Φ AR θ AR =w

[0074] in, is the AR model coefficient, and the estimated value is calculated using the least squares method

[0075]

[0076] According to the Yule-Walker equation, we can further obtain the variance estimate of the input signal {w(n,m)}, that is, w

[0077]

[0078] Step 8: Substitute the estimated random input signal Substitute the original ARMA model expression to solve the parameter estimation value

[0079] Get an estimate of the noise After that, we can get the estimated value of the matrix

[0080]

[0081] If it is known, the model can be solved according to the least squares method to obtain the least squares estimate of the parameter vector of the original ARMA model. The least squares formula is as follows

[0082]

[0083] Step 9: Repeat steps 7 and 8 for the R, G, and B channel images of the tongue image to obtain That is, the individual MCMD ARMA model is used to describe the human body system.

[0084] The present application embodiment provides a method for evaluating the performance of a human body system modeled by a MCMD ARMA model. For an object to be measured,

[0085] After completing the modeling of the human body system with the MCMD ARMA model, the response of the human body to changes in the external environment can be simulated under a given input signal, that is, the estimated Gaussian random signal is used as the input of the LTI human body system, and the reconstructed output signal is obtained through the response of the LTI system. The accuracy of the modeling of the human body system is evaluated by comparing the difference between the reconstructed signal of the model and the real signal obtained by actual measurement, that is, the clinical medical data.

[0086] Step 1: Use the AIC criterion to determine the model order, and then use the least squares method to estimate the coefficients and random input signals of the MCMD ARMA model to obtain an individual mathematical model used to describe the LTI human system.

[0087] Step 2: The estimated Gaussian random signal is used to represent the influence of the external environment on the human body as the input of the LTI human body system, that is, the signal estimation value can be obtained for each single-channel image data The output signal estimate of the LTI human system can be obtained The model process simulates the response process of the human body to the complex and changing external environment. The estimated output signal is used as an approximation of the actual measured clinical medical signal. The accuracy of the model in modeling the human body is evaluated by comparing the difference between the model output signal and the actual measured data of the human body.

[0088] Step 3: Use the normalized root mean square error (NRMSE) as a performance indicator to objectively evaluate the accuracy of the MCMD ARMA model in modeling the human system. And calculate the average of the NRMSE of each channel of all images. The NRMSE formula is as follows

[0089]

[0090] in is the predicted value, is the true value, mean(·) is the mean, max(·) and min(·) represent the maximum and minimum values ​​respectively, ‖·‖ 2 It means to find the second norm.

[0091] This application embodiment provides a possible implementation method as follows:

[0092] The simulation results of human body system modeling and performance evaluation based on MCMD ARMA model using tongue image signals are as follows: Figure 3 This is a comparison diagram of the tongue image reconstruction signal and the real measurement signal obtained after the selected experimental object is modeled by the MCMD ARMA model in the embodiment of this application. The experimental scene is set as follows:

[0093] Volunteers willing to cooperate with the research project were selected at the hospital and their tongue images were taken. Figure 5 This is a tongue image clinical data measurement diagram provided in an embodiment of the present application.

[0094] The camera model used in this application example is Canon EOS RP, with a fixed focal length. The data acquisition light source adopts a 50D standard light source. A semi-enclosed acquisition environment is designed to reduce the impact of ambient light on the captured image, providing stable and standardized data acquisition conditions.

[0095] The tongue image segmentation method is used to extract the tongue from irrelevant background to achieve data preprocessing. The present application embodiment uses the U-Net network to achieve tongue image segmentation, and the effect diagram is as follows: Figure 6 shown.

[0096] The order of the model is determined to be p based on the AIC method. 1 =p 2 =q 1 =q 2 =3, establish 3C2D ARMA (3,3,3,3), that is, for each single channel image of R, G, and B, establish a 2D ARMA (3,3,3,3) model, and use the least squares method to estimate the model coefficient matrix. The parameter AR matrix A of the three channel images of R, G, and B can be obtained respectively. 1 ,A 2 ,A3 and MA parameter matrix B 1 ,B 2 ,B 3 , that is, the estimated parameters Complete the modeling of this human system.

[0097] The tongue image signal was reconstructed using the trained 3C2D ARMA (3,3,3,3) model and the NRMSE was calculated. This application example collected data from 230 volunteers in total, and the average NRMSE results were displayed in a table. Figure 4 middle.

Claims

1. A human body system modeling method based on a MCMD ARMA model of a tongue image, the method comprising: Step 1: Collect clinical data and place corresponding data collection equipment; Step 2: Obtain the sample information received by the data acquisition device, which is the relevant clinical medical information of the subject, and obtain the tensor data of M-dimensional K channels where L1,…,L M express The size in each dimension, K represents the number of channels; Step 3: MCMD ARMA model based on multidimensional and multichannel image data; MCMD stands for multidimensional and multichannel, and ARMA stands for autoregressive moving average; For M-dimensional K-channel tensor data Divide the model into equal-sized blocks according to its AR order As mentioned above, where I1,…,I M express The size in each dimension, K represents the number of channels, x ′ (i1,…,i M ,k) represents a tensor At the point (i1,i2…,i M ,k) element value, AR represents the autoregressive model; Similarly, for the original data Consider noise tensors of equal size Divide the MA into noise tensor blocks according to its order where J1,…,J N express The size in each dimension, K represents the channel, and MA represents the self-sliding average model; Suppose an ARMA system with M dimensions and K channels, defined as follows: in, and are the coefficient tensors of the AR and MA parts respectively; Represents the AR parameter tensor and data tensor The tensor contraction product of the first M dimensions of is a second-order tensor, i.e., a K×K matrix, with matrix element index express Tensor parameters for channels and The data tensor of the channel The tensor contraction product of is expressed as The formula is as follows: Similarly, Represents the MA parameter tensor and the noise tensor The tensor contraction product of the first N dimensions is also a K×K matrix; Define the AR part of the model as The MA part is The expression is written as: in Symbols represent definitions; Step 4: Convert the MCMD ARMA model into multiple single-channel multidimensional ARMA models; Slice the data according to different channels. A data tensor with K different channels constitute; AR parameter tensor consisting of K different channels constitute, MA parameter tensor consisting of K different channels constitute, A noise tensor consisting of K different channels Construct; write the original expression in matrix form: Since the MCMD ARMA model is transformed into multiple single-channel multidimensional ARMA models, it is defined as Then the corresponding matrix elements of the above expression are equal, and the M-dimensional K-channel ARMA model corresponds to K×K single-channel sub-models, that is, the K×K single-channel M-dimensional ARMA model is: Where 1≤i,j≤K, The <*,*> operator represents the dot product sum of the corresponding positions of two data of the same size; Step 5: Design optimization function solution model based on different experimental requirements; The residual ε of the ARMA model with M dimensions and K channels is defined as: Among them, e ij ,1≤i,j≤K corresponding model The residual of , that is, the parameter tensor of channel i The data tensor corresponding to channel j and the noise tensor The single-channel ARMA model of , the following discusses the two cases of channel coherence and incoherence respectively; If the correlation between different channels needs to be considered, the corresponding residuals are weighted and summed; for the parameter tensor of the i-th channel 1≤i≤K, the ARMA model optimization problem of the M-dimensional K-channel is: where g ij ,1≤i,j≤K is the weighting coefficient. Different weighting coefficients are set according to the coherence of each channel. That is, the ARMA model residual ε of M-dimensional K-channel is optimized by the weighting coefficient matrix G. The expression of G is as follows: Different ARMA models reflect the coherence of data from different channels, such as the sub-model Then the correlation between the data of channel 1 and channel K is measured; If channel correlation does not need to be considered, the weighted coefficient matrix G of the ARMA model with M-dimensional K channels is a unit matrix, as shown below: The optimization problem is: Only the diagonal elements of the model residual matrix need to be optimized to convert the original M-dimensional K-channel ARMA model into K×K single-channel sub-models. Convert to the following K multi-dimensional single-channel model, that is, the M-dimensional single-channel ARMA model for solution: Where 1≤i≤K, Step 6: Determine the model order AIC through Akaike information; AIC=-2*ln(L)+2*k Among them, L is the likelihood function value of the maximum likelihood estimate of the model, and k is the number of parameters to be estimated; Step 7: Modeling of the human body system with tongue image as system output based on MCMD ARMA model; For an RGB tongue image data of size L1×L2, that is, two-dimensional three-channel data x(i1,i2,i3) represents the pixel value at point (i1,i2,i3); i1,i2 are position indexes, with the upper left corner of the image as the coordinate origin; i3 represents the channel index, when i3=1, 2, 3, it represents R, G, B channels respectively; similarly, consider that each image has a noise image of equal size, that is, have Therefore, the noise input of the ARMA system is w(j1,j2,j3) represents the noise value at point (j1,j2,j3), where j1 and j2 are position indexes, with the upper left corner of the noise image as the origin of the coordinates; j3 represents the channel index, and when j3=1, 2, 3, it represents the R, G, and B channels respectively; Tongue image data At the image reference points x(n,m,1), x(n,m,2) and x(n,m,3), where n and m represent the position index, a sliding window of size (p1+1)×(p2+1)×3 is used to take values ​​in the 1 / 4 support area of ​​the image to obtain the image data block. The same is true for noisy data According to the sliding window of size (q1+1)×(q2+1)×3, the value is taken in the 1 / 4 support area, and the noise tensor data block can be obtained. Where p1, p2, q1, q2 are two-dimensional ARMA orders, max{p1, q1}<n≤L1-1, max{p2, q2}<m≤L2-1; tongue image data block and noise block The equations in the AR and MA parts are as follows: in, and are the coefficient tensors for the AR and MA parts of the model respectively; AR parameter tensor According to different channel slices, we can get Similarly, MA parameter tensor According to different channel slices, we can get Will and Written in matrix form: in, and The coefficient tensors A1, A2, and A3 for the AR and MA parts of the model are respectively composed of the model AR parameter tensor The front slice is obtained, B1, B2, B3 are the model MA parameter tensors The front slice is obtained, 1≤i,j≤3, A i ,B j The expression is as follows: Since the tongue image data is two-dimensional three-channel data, and The nine components correspond to nine two-dimensional ARMA systems, namely 2D ARMA systems, and the model residuals The expression is: Since the R, G, and B channels are orthogonal, for The calculated value does not need to consider the coherence between channels. The optimization expression is min(tr(ε)), where tr(·) is the trace of the matrix, that is, tr(ε) = e 11 +e 22 +e 33 ; The following three single-channel 2D ARMA models need to be optimized <A1,X1> =<B1,W1> <A2,X2> =<B2,W2> <A3,X3> =<B3,W3> ; Step 8: For each single-channel image of the tongue image, approximate the 2D ARMA model with the 2D AR model; because<A1,X1> =<B1,W1> ,<A2,X2> =<B2,W2> and<A3,X3> =<B3,W3> The model parameter solution process is the same as the model<A1,X1> =<B1,W1> For example, all subsequent derivations are based on the model<A1,X1> =<B1,W1> , does not involve other channel parameters. For the convenience of writing, the channel item parameters are omitted and expressed as<A,X> =<B,W> ; For a single-channel two-dimensional image data X = {x(i1, i2): 0≤i1≤L1-1, 0≤i2≤L2-1}, consider the noise map of equal size W = {w(j1, j2): 0≤j1≤L1-1, 0≤j2≤L2-1}; the ARMA model at the image point x(i1, i2) is expressed as: in θ is the model parameter vector to be solved, which is defined as follows: Same as step 7, p1, p2, q1, q2 are two-dimensional ARMA orders; Since in image X, the equation is valid for all pixels x(n,m) that satisfy n=P+1,...,L1-1; m=Q+1,...,L2-1, P>max(p1,q1),Q>max(p2,q2), we can expand it into matrix form and get x+Φθ=w Among them, x is the output signal vector, θ is the model parameter vector to be solved, and w is the input signal vector, that is, the noise vector. The expression is: x=[x(P+1,Q+1),x(P+2,Q+1)…,x(L1-1,L2-1)] T w=[w(P+1,Q+1),w(P+2,Q+1)…,w(L1-1,L2-1)] T Since the model noise term W is unknown, there are unknown terms in Φ that cannot be solved directly. In order to solve the model input term W, the AR(K1, K2) model is used to approximate the ARMA model. The AR(K1, K2) model expression established at x(n, m) is: a ij ′ represents the AR(K1,K2) model parameters; According to the Yule-Walker equation, we can get Assume a 00 =1,σ 2 is the variance of the noise term W, δ(k,l) is the two-dimensional Kronecker function, and r(k,l) is the autocorrelation value, which is obtained through the original image data and is defined as follows: r(k,l)=r(-k,-l),(k,l)≥(0,0) r(-k,l)=r(k,-l),(k,l)≥(1,1) Get the AR(K1,K2) model parameter a ij ′ Estimated value Similarly, the equation is valid for the original image pixel x(n,m) that satisfies the conditions n=K1,...,L1-1; m=K2,...,L2-1, and it can be expanded into a matrix form to obtain: x+Φ AR i AR =w Among them, Φ AR The expression is: and is the AR model coefficient; According to the Yule-Walker equation, the estimated value is calculated using the least squares method. The estimated error of the model is Its variance is the variance estimate of the noise term W The noise term in the experiment is Gaussian white noise, and the corresponding noise can be generated according to As input; Step 9: Substitute the estimated random input signal Substitute into the ARMA model expression and solve for AR parameters A1, A2, A3 and MA parameters B1, B2, B3; Step 10: Repeat steps 7 and 8 for the R, G, and B channel images of the tongue image to obtain That is, the individual 3C2D ARMA model is used to describe the human body system.

2. The human body system modeling method based on the MCMD ARMA model of the tongue image according to claim 1, characterized in that: The specific method of step 9 is: Get an estimate of the noise After that, we get the estimated value of the matrix If it is known, the model is solved according to the least squares method to obtain the least squares estimate of the parameter vector of the original ARMA model. The least squares formula is as follows:

3. A method for evaluating a human body model established by the method of claim 1, the method comprising: Step 1: Use the MCMD ARMA model to model the human body system with tongue images as output; A 3C2D ARMA model was established based on the RGB tongue image. Considering that there is no coherence between channels, it was converted into three 2D ARMA models. The Akaike information criterion was used to determine the model order, and then the least squares method was used to estimate the coefficients of the model and the random input signal to obtain an individual mathematical model for describing the human body system. Step 2: The generated random signal that obeys Gaussian distribution is used to represent the impact of external random environmental factors on the human body as the input of the human body system, that is, Based on this input signal and the estimated parameters Calculate output signals of human body systems The estimated output signal is regarded as an approximation of the actual measured clinical medical signal, and the accuracy of the model in modeling the human body is evaluated by comparing the difference between the model's output signal and the real measured data; The normalized root mean square error is used as a performance indicator to objectively evaluate the accuracy of the model in modeling the human system.

Images