Method, system and electronic device for constructing functional connection network

Abstract

The present invention discloses a method for constructing a functional connectivity network, a system and an electronic device, including: using a variational autoencoder network to establish a multivariate Gaussian distribution probability model for each brain region of interest through unsupervised learning, using Jensen-Shannon divergence to calculate the degree of similarity between the probability distributions of any two brain regions, and obtaining an adjacency matrix of paired brain regions as a brain functional network; using a two-sample t-test to select a feature set with a higher discrimination, and inputting the selected feature set into a support vector machine for classification; inputting the feature sets of different dimensional models into a support vector machine classifier, and using adaboost enhancement method to combine multiple classifiers to obtain a stronger classifier. The present invention uses a generative learning method to improve the statistical power of group data and the generalization ability of observation data variability in the diagnosis of mild cognitive impairment, and further improves the diagnostic accuracy of mild cognitive impairment through the joint enhancement of multidimensional models.

Description

Technical Field

[0001] The present invention relates to the technical fields of image processing and pattern recognition, and in particular to a method, system and electronic equipment for constructing a functional connection network. Background Art

[0002] Alzheimer's disease (AD) is an irreversible, chronic neurodegenerative disease characterized by progressive decline in cognitive and memory functions. It develops slowly and is often not easily detected.

[0003] Mild cognitive impairment (MCI) is a condition between healthy individuals and Alzheimer's disease (AD), and is generally considered the prodromal stage of Alzheimer's disease (AD). Correctly diagnosing MCI not only improves the quality of life of the elderly but also allows for early prevention of AD.

[0004] In order to improve the accuracy of MCI diagnosis, it is generally necessary to construct an efficient functional connectivity network (FCN). Traditional methods for constructing functional connectivity networks (FCN), such as Pearson correlation and sparse representation methods, have greatly promoted the development of FCN estimation models and improved the accuracy of MCI diagnosis. However, since the observed data (e.g., blood oxygenation level-dependent (BOLD) time series) contain complex noise, human factors, and individual differences between different subjects, how to construct a biologically meaningful and robust functional connectivity network (FCN) remains a challenging problem. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to use a limited amount of rs-fMRI data containing complex noise to construct a functional connection network with strong generalization ability. A method, system and electronic device for constructing a functional connection network are provided to solve the above problem.

[0006] A first object of the present invention is to provide a method for constructing a functional connectivity network, comprising:

[0007] S1: rs-fMRI data of mild cognitive impairment and normal control groups that meet the criteria were selected from the ADNI database;

[0008] S2: Preprocess the selected rs-fMRI data and extract the average time series of the region of interest using the AAL template;

[0009] S3: Constructing a brain functional network based on generative learning: Gaussian distributions of different dimensions of each brain region are obtained through generative network VAE. Jensen-Shannon divergence is used to calculate the similarity between the probability distributions of any two brain regions, and the adjacency matrix of paired brain regions is obtained.

[0010] S4: A two-sample t-test was used to select a feature set with high discrimination for the obtained brain functional network;

[0011] S5: Use adaboost enhancement method to combine the support vector machine classifiers trained with feature sets of different dimensional models to obtain a strong classifier, which is used to classify and diagnose mild cognitive impairment (MCI).

[0012] Furthermore, the data preprocessing in step S2 includes: removing time points, time layer correction, head motion correction, spatial normalization, spatial smoothing, removing linear drift, filtering, removing covariates and extracting regions of interest.

[0013] Furthermore, the data preprocessing in step S2 specifically includes:

[0014] Step S 21 : The first 10 rs-fMRI images of each subject were discarded;

[0015] Step S 22 : Correct the remaining images to eliminate the influence of head motion;

[0016] Step S 23 : Spatial normalization, smoothing, and bandpass filtering between 0.01 and 0.1 Hz;

[0017] Step S 24 : Regress the cerebrospinal fluid signal and white matter, and eliminate subjects with frame-wise displacement greater than 0.5 mm;

[0018] Step S 25 : The BOLD time series signal was divided into 90 regions of interest (ROIs) based on the AAL template.

[0019] Furthermore, the step S3 includes:

[0020] Step S 31 : Train the generative network VAE in an unsupervised learning manner;

[0021] Step S 32 : Use the encoder part of the trained generative network VAE to obtain the mean and covariance of the multivariate Gaussian function of the corresponding brain area;

[0022] Let P1 and P2 be the probability density functions of two regions of interest (ROIs), both of which are independent multivariate Gaussian distributions obtained by the generative network VAE, then:

[0023] P1=N(x|μ1,∑1)

[0024] P2=N(x|μ2,∑2)

[0025] Where: x is the input of the generative network VAE, that is, the average time series of the region of interest (ROI), u1 is the mean of the multivariate Gaussian function of the first brain region, Σ1 is the covariance of the multivariate Gaussian function of the first brain region, u2 is the mean of the multivariate Gaussian function of the second brain region, Σ2 is the covariance of the multivariate Gaussian function of the second brain region, and N(.) represents the multivariate Gaussian distribution based on the region of interest (ROI);

[0026] Step S 33 : Jensen-Shannon divergence was used to measure the statistical similarity between the probability density functions P1 and P2 of two regions of interest (ROIs) and to estimate the functional connectivity matrix of the brain regions.

[0027] Furthermore, the step S 31 Specifically:

[0028] The average time series of the region of interest (ROI) is used as input and the encoder is used to obtain the multivariate Gaussian distribution in the latent space.

[0029] Sampling a vector from the latent space using a reparameterized approach;

[0030] The vector is fed into the decoder to reconstruct the original data or generate new data;

[0031] The loss function of the generative network VAE consists of two parts, the reconstruction error and the regularization term that makes the latent space regular, which is expressed as follows:

[0032]

[0033] Among them, x is the input of the generative network VAE, that is, the average time series of the region of interest ROI; is the output of the generative network, i.e. the result of reconstruction; parameter μ x Represents the average mean value input to the generating network VAE, σ x Represents the covariance of the input to the generating network VAE, function D KL () represents the regularization function that makes the latent space regular, and N() represents the multivariate Gaussian distribution.

[0034] Furthermore, the step S 32 In the equation, the probability density function of the region of interest ROI is expressed as:

[0035]

[0036] Where n is the dimension of the Gaussian distribution, μ is the mean vector, and ∑ is the diagonal element. The covariance matrix of , x is the average time series of ROI, and T represents the transpose.

[0037] Furthermore, in step S 33 In the above equation, the Jensen-Shannon divergence is a variant of the Kullback-Leibler divergence and is defined as follows:

[0038]

[0039] The Kullback-Leibler divergence is represented by the means and covariance matrices of two distributions, and is defined as follows:

[0040]

[0041] Where tr(·) represents the trace of the matrix, D JS is the Jensen-Shannon divergence, D KL is the Kullback-Leibler divergence.

[0042] Furthermore, the feature set selection in step S4 includes:

[0043] Step S 41 :The upper triangular elements of the relationship matrix of the brain functional network are selected as input features for identifying mild cognitive impairment (MCI);

[0044] Step S 42 : A two-sample t-test with a significance level of p < 0.01 was used to select a feature set with high discrimination.

[0045] Furthermore, the step S5 includes:

[0046] Step S 51 : Using the adaboost enhancement method to combine the support vector machine classifiers trained on the feature sets of different dimensional models to obtain a strong classifier;

[0047] Step S 52 : The strong classifier is used to classify and diagnose mild cognitive impairment (MCI).

[0048] A second object of the present invention is to provide a system for constructing a functional connection network, comprising:

[0049] Data acquisition module, used to obtain rs-fMRI data of mild cognitive impairment and normal control groups in the ADNI database;

[0050] Data processing module, used to preprocess rs-fMRI data and extract the average time series of the region of interest (ROI) using the AAL template;

[0051] Network building module, used to construct brain functional networks based on generative learning;

[0052] The feature determination module is used to determine the feature set with high discrimination for the brain functional network generated by the VAE of the generative network with mild cognitive impairment in different dimensional latent spaces, and use adaboost to train the feature sets of different dimensional models to obtain SVM;

[0053] The obstacle recognition module is used to input the test set into the strong classifier to obtain the MCI classification result to determine the user's cognitive impairment recognition result.

[0054] A third object of the present invention is an electronic device comprising: one or more processors;

[0055] Memory;

[0056] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method for constructing a functional connection network as described above.

[0057] The fourth object of the present invention is a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for constructing a functional connection network.

[0058] The beneficial effects brought about by the technical solution provided by the present invention are:

[0059] (1) The present invention encodes the average time series of the region of interest (ROI) into a low-dimensional space using a generative network (VAE), so that the obtained brain functional network has the key characteristics of low noise and rich information;

[0060] (2) Generative learning is used to map each average time series into a probability distribution in the latent space, which improves the statistical power of group data and the generalization ability of observational data variability in the diagnosis of mild cognitive impairment;

[0061] (3) The feature sets of different dimensional models are input into the SVM classifier, and multiple classifiers are combined using the adaboost enhancement method to obtain a stronger classifier, further improving the diagnostic accuracy of mild cognitive impairment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The same reference numerals are used in these drawings to represent the same or similar parts. The exemplary embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0063] Figure 1 This is a flowchart of the method for constructing a functional connection network of the present invention;

[0064] Figure 2 A flowchart of the present invention for modeling brain functional networks based on generative learning;

[0065] Figure 3 This is a flowchart of the MCI classification based on the brain functional network of generative learning in the present invention;

[0066] Figure 4 This is a structural diagram of a functional connection network construction system in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the technical features, purposes and effects of the present invention more clearly understood, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0068] like Figure 1 As shown, the present invention proposes a method for constructing a functional connectivity network, comprising the following steps:

[0069] S1: Resting-state functional magnetic resonance imaging (rs-fMRI) data of mild cognitive impairment and normal control groups that met the criteria were selected from the ADNI database;

[0070] It should be noted here that the ADNI database is a public data platform focusing on neuroimaging of Alzheimer's disease. In addition to imaging resources such as MRI (sMRI / fMRI / DTI) and Aβ-PET, it also covers clinical information such as the subjects' body fluid markers (blood, cerebrospinal fluid samples), neuropsychiatric scales, etc.

[0071] S2: Preprocess the selected resting-state functional magnetic resonance imaging (rs-fMRI) data and extract the average time series of the region of interest using the AAL template;

[0072] The AAL template, also known as the Anatomical Automatic Labeling (AAL) template, is provided by the Montreal Neurological Institute (MNI). Currently, the AAL template includes 116 regions, but only 90 of them belong to the cerebrum, with the remaining 26 belonging to the cerebellum.

[0073] Specifically, in an embodiment of the present invention, the data preprocessing in step S2 includes: removing time points, time layer correction, head motion correction, spatial normalization, spatial smoothing, removing linear drift, filtering, removing covariates and extracting regions of interest.

[0074] To ensure signal stability, data preprocessing specifically includes:

[0075] Step S 21 : The first 10 rs-fMRI images of each subject were discarded;

[0076] Step S 22 : Correct the remaining images to eliminate the influence of head motion;

[0077] Step S 23 : Spatial normalization, smoothing, and bandpass filtering between 0.01 and 0.1 Hz;

[0078] Step S 24 : Regress the cerebrospinal fluid signal and white matter, and eliminate subjects with frame-wise displacement greater than 0.5 mm;

[0079] Step S 25 : The blood oxygen level dependent (BOLD) time series signal was divided into 90 regions of interest (ROIs) based on the AAL template.

[0080] S3: Constructing a brain functional network based on generative learning: Gaussian distributions of different dimensions of each brain region are obtained through generative network VAE. The similarity between the probability distributions of any two brain regions is calculated using Jensen-Shannon divergence. The adjacency matrix of paired brain regions is obtained, which is the brain functional network.

[0081] To generate a network VAE model, please refer to Figure 2 As shown in Figure 2, a variational autoencoder (VAE) is an autoencoder whose encoding distribution is normalized during training to ensure that it has good properties in the latent space, allowing us to generate some new data.

[0082] Specifically, in the embodiment of the present invention, step S3 specifically includes:

[0083] Step S 31 The VAE network is trained in an unsupervised manner. It takes the average time series of the ROI as input and passes it through the encoder to obtain a multivariate Gaussian distribution in the latent space. A vector is sampled from this latent space using the reparameterization technique and input into the decoder to reconstruct the original data or generate new data.

[0084] The loss function of the generative network VAE consists of two parts: the reconstruction error and a regularization term that regularizes the latent space, as shown below:

[0085]

[0086] Among them, x is the input of the generative network VAE, that is, the average time series of the region of interest ROI; is the output of the generative network, i.e. the result of reconstruction; parameter μ x Represents the average mean value input to the generating network VAE, σ x Represents the covariance of the input to the generating network VAE, function D KL () represents the regularization function that makes the latent space regular, and N() represents the multivariate Gaussian distribution.

[0087] Regularization of the latent space is achieved by making the probability distribution in the latent space approximate the standard normal distribution.

[0088] Step S 32 : In the stage of constructing the brain functional network, only the encoder part of the trained generative network VAE is used to obtain the mean and covariance of the multivariate Gaussian function of the corresponding brain area.

[0089] Assume that P1 and P2 are the probability density functions of two regions of interest (ROIs), both of which are independent multivariate Gaussian distributions obtained by the generative network VAE, that is, P1 = N(x|μ1,∑1) and P2 = N(x|μ2,∑2). The probability density function based on the region of interest (ROI) is expressed as follows:

[0090]

[0091] In the above formula, n is the dimension of Gaussian distribution, μ is the mean vector, and ∑ is the diagonal element. The covariance matrix of , x is the average time series of ROI, and T represents the transpose.

[0092] Step S 33 : The Jensen-Shannon divergence is a variant of the Kullback-Leibler divergence, which is often used to measure the statistical similarity between two probability density functions. The Kullback-Leibler divergence of two normal distributions can be represented by the mean and covariance matrix of the two distributions as follows:

[0093]

[0094] where tr(·) represents the trace of the matrix.

[0095] Since the Kullback-Leibler divergence is usually asymmetric, the present invention uses a variant of the Kullback-Leibler divergence, namely the Jensen-Shannon divergence, to estimate the functional adjacency matrix of brain regions, which is defined as follows:

[0096]

[0097] Among them, D JS is the Jensen-Shannon divergence, D KL is the Kullback-Leibler divergence.

[0098] It's worth noting that the Jensen-Shannon divergence measures the similarity between two probability distributions. It's a variant of the KL divergence that addresses the asymmetric nature of the KL divergence. Generally, the JS divergence is symmetric, ranging from 0 to 1.

[0099] S4: A two-sample t-test was used to select a feature set with higher discrimination for the obtained brain functional network;

[0100] S5: Use the adaboost enhancement method to combine the support vector machine classifiers trained on the feature sets of different dimensional models to obtain a strong classifier, and use this strong classifier to classify and diagnose MCI.

[0101] AdaBoost (adaptive boosting) is an adaptive enhancement method that can automatically adjust the sample weight according to the error rate of the previous base model's sample prediction, and then continue to learn a new base model based on the samples with adjusted weights. This process is repeated until the number of base models reaches the set number. Finally, all the trained base models are integrated through a combination strategy to obtain the final model.

[0102] like Figure 3 As shown in the figure, since the relationship matrix of the brain functional network is symmetrical, only its upper triangular elements are considered as input features for identifying MCI, that is, the dimension of the feature vector is 90×(90-1) / 2=4005. The feature dimension is very high relative to the number of samples. A high-dimensional feature set not only increases the computational complexity, but may also affect the classification performance.

[0103] Therefore, the present invention uses a two-sample t-test with a significance level of p < 0.01 to select a feature set with high discrimination. Then, the feature sets of different dimensional models are used to train a support vector machine with a linear kernel. Since the feature sets generated by the VAE network with different dimensional latent spaces have good complementary properties in terms of precision and recall for the diagnosis of MCI,

[0104] Therefore, the present invention uses the adaboost enhancement method to combine multiple classifiers to obtain a stronger classifier, and then inputs the test set into the strong classifier to obtain the MCI classification result, thereby improving the accuracy of MCI diagnosis.

[0105] In addition, the so-called latent space specifically refers to the space from the initial space to the encoding space, that is, the encoder compressed data.

[0106] See also Figure 4 As shown, an embodiment of the present invention further provides a system for constructing a functional connection network, including:

[0107] Data acquisition module, used to obtain rs-fMRI data of mild cognitive impairment and normal control groups in the ADNI database;

[0108] Data processing module, used to preprocess rs-fMRI data and extract the average time series of the region of interest (ROI) using the AAL template;

[0109] Network building module, used to construct brain functional networks based on generative learning;

[0110] The feature determination module is used to determine the high-discrimination feature set of the brain functional network generated by the VAE generative network of different dimensional latent spaces, and use the adaptive enhancement method adaboost to train the feature sets of different dimensional models to obtain the support vector machine (SVM);

[0111] The obstacle recognition module is used to input the test set into the strong classifier to obtain the classification result of mild cognitive impairment (MCI) to determine the user's cognitive impairment recognition result.

[0112] An embodiment of the present invention further provides an electronic device, comprising: one or more processors;

[0113] Memory;

[0114] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method for constructing a functional connection network as described above.

[0115] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the above-mentioned method for constructing a functional connection network.

[0116] Therefore, the present invention solves the problem of how to use a limited amount of rs-fMRI data containing complex noise to construct an FCN with strong generalization ability by establishing a model of brain functional network based on generative learning.

[0117] First, a multivariate Gaussian distribution probability model is established for each region of interest using a generative network (VAE) through unsupervised learning. The similarity between the probability distributions of any two brain regions is calculated using the JS divergence, and the adjacency matrix of paired brain regions is obtained, which is the brain functional network.

[0118] Then, in order to reduce the dimension of the input features and reduce the interference of insignificant features, a two-sample t-test is used to select a feature set with higher discrimination, making the selected features more directional;

[0119] Finally, multiple SVM classifiers were combined through adaboost enhancement to obtain a stronger classifier, which was used to classify MCI, thereby improving the accuracy of MCI diagnosis.

[0120] The examples of the present invention are described above in conjunction with the accompanying drawings, but the present invention does not take precedence over the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A method for constructing a functional connectivity network, characterized in that: include: S1: rs-fMRI data of mild cognitive impairment and normal control groups that meet the criteria were selected from the ADNI database; S2: Preprocess the selected rs-fMRI data and extract the average time series of the region of interest using the AAL template; S3: Construct a brain functional network based on generative learning. Use the generative network VAE to obtain Gaussian distributions of different dimensions for each brain region. Use Jensen-Shannon divergence to calculate the similarity between the probability distributions of any two brain regions and obtain the adjacency matrix of paired brain regions. Specifically include: Step S 31 : Train the generative network VAE in an unsupervised learning manner; The step S 31 Specifically: The average time series of the region of interest (ROI) is used as input and the encoder is used to obtain the multivariate Gaussian distribution in the latent space. Sampling a vector from the latent space using a reparameterized approach; Inputting the vector into a decoder to reconstruct the original data or generate new data; The loss function of the generative network VAE consists of two parts, the reconstruction error and the regularization term that makes the latent space regular, which is expressed as follows: in, It is the input of the generative network VAE, that is, the average time series of the region of interest ROI; is the output of the generative network, i.e. the result of reconstruction; parameter μ x Represents the average mean value input to the generating network VAE, σ x Represents the covariance of the input to the generating network VAE, function D KL () represents the regularization function that makes the latent space regular, N() represents the multivariate Gaussian distribution; Step S 32 : Use the trained VAE encoder to obtain the mean and covariance of the multivariate Gaussian function of the corresponding brain region; make and are the probability density functions of the two regions of interest (ROIs), both of which are independent multivariate Gaussian distributions obtained by the generative network VAE, then: in: is the input of the generative network VAE, that is, the average time series of the region of interest ROI, is the mean of the multivariate Gaussian function of the first brain area, is the covariance of the multivariate Gaussian function of the first brain area, is the mean of the multivariate Gaussian function of the second brain area, is the covariance of the multivariate Gaussian function of the second brain area, Represents a multivariate Gaussian distribution based on the region of interest (ROI); The expression of the probability density function of the region of interest ROI is: in, is the dimension of the Gaussian distribution, is the mean vector, The diagonal elements are The covariance matrix of , x is the average time series of ROI, T represents the transpose; Step S 33 :Use Jensen-Shannon divergence to measure the probability density functions of two regions of interest ROI and The statistical similarity between the two is used to estimate the functional connectivity matrix of the brain region; the Jensen-Shannon divergence is a variant of the Kullback-Leibler divergence, and the Jensen-Shannon divergence is defined as follows: The Kullback-Leibler divergence is represented by the means and covariance matrices of two distributions, and is defined as follows: where tr(·) represents the trace of the matrix, is the Jensen-Shannon divergence, D KL is the Kullback-Leibler divergence; S4: A two-sample t-test was used to select a feature set with higher discrimination for the obtained brain functional network; S5: Using the adaboost enhancement method, support vector machine classifiers trained on feature sets of different dimensional models are combined to obtain a strong classifier, and the strong classifier is used to classify and diagnose mild cognitive impairment (MCI).

2. The method for constructing a functional connectivity network according to claim 1, wherein: The preprocessing of the selected rs-fMRI data includes: removing time points, time layer correction, head motion correction, spatial normalization, spatial smoothing, removing linear drift, filtering, removing covariates and extracting regions of interest.

3. The method for constructing a functional connectivity network according to claim 2, wherein: In step S2, data preprocessing specifically includes: Step S 21 : The first 10 rs-fMRI images of each subject were discarded; Step S 22 : Correct the remaining images to eliminate the influence of head motion; Step S 23 : Spatial normalization, smoothing, and bandpass filtering between 0.01 and 0.1 Hz; Step S 24 : Regress the cerebrospinal fluid signal and white matter, and remove subjects with frame-wise displacement greater than 0.5 mm; Step S 25 : The BOLD time series signal was divided into 90 regions of interest (ROIs) based on the AAL template.

4. The method for constructing a functional connectivity network according to claim 1, wherein: In step S4, the selection of the feature set specifically includes: Step S 41 :The upper triangular elements of the relationship matrix of the brain functional network are selected as input features for identifying mild cognitive impairment (MCI); Step S 42 : A two-sample t-test with a significance level of p < 0.01 was used to select feature sets with higher discrimination.

5. The method for constructing a functional connectivity network according to claim 1, wherein: In step S5, it specifically includes: Step S 51 : Using the adaboost enhancement method to combine the support vector machine classifiers trained on the feature sets of different dimensional models to obtain a strong classifier; Step S 52 : The strong classifier is used to classify and diagnose mild cognitive impairment (MCI).

6. A system for constructing a functional connectivity network, characterized in that: include: Data acquisition module, used to obtain rs-fMRI data of mild cognitive impairment and normal control groups in the ADNI database; Data processing module, used to preprocess rs-fMRI data and extract the average time series of the region of interest (ROI) using the AAL template; The network construction module is used to construct a brain functional network based on generative learning. The Gaussian distribution of different dimensions of each brain region is obtained by generating a network VAE. The Jensen-Shannon divergence is used to calculate the similarity between the probability distributions of any two brain regions, and the adjacency matrix of paired brain regions is obtained. Specifically including step S 31 : Taking the average time series of the region of interest (ROI) as input, the encoder obtains the multivariate Gaussian distribution in the latent space; Sampling a vector from the latent space using a reparameterized approach; Inputting the vector into a decoder to reconstruct the original data or generate new data; The loss function of the generative network VAE consists of two parts, the reconstruction error and the regularization term that makes the latent space regular, which is expressed as follows: in, It is the input of the generative network VAE, that is, the average time series of the region of interest ROI; is the output of the generative network, i.e. the result of reconstruction; parameter μ x Represents the average mean value input to the generating network VAE, σ x Represents the covariance of the input to the generating network VAE, function D KL () represents the regularization function that makes the latent space regular, N() represents the multivariate Gaussian distribution; Step S 32 : Use the trained VAE encoder to obtain the mean and covariance of the multivariate Gaussian function of the corresponding brain region; make and are the probability density functions of the two regions of interest (ROIs), both of which are independent multivariate Gaussian distributions obtained by the generative network VAE, then: in: is the input of the generative network VAE, that is, the average time series of the region of interest ROI, is the mean of the multivariate Gaussian function of the first brain area, is the covariance of the multivariate Gaussian function of the first brain area, is the mean of the multivariate Gaussian function of the second brain area, is the covariance of the multivariate Gaussian function of the second brain area, Represents a multivariate Gaussian distribution based on the region of interest (ROI); The expression of the probability density function of the region of interest ROI is: in, is the dimension of the Gaussian distribution, is the mean vector, The diagonal elements are The covariance matrix of , x is the average time series of ROI, T represents the transpose; Step S 33 :Use Jensen-Shannon divergence to measure the probability density functions of two regions of interest ROI and The statistical similarity between the two is used to estimate the functional connectivity matrix of the brain region; the Jensen-Shannon divergence is a variant of the Kullback-Leibler divergence, and the Jensen-Shannon divergence is defined as follows: The Kullback-Leibler divergence is represented by the means and covariance matrices of two distributions, and is defined as follows: where tr(·) represents the trace of the matrix, is the Jensen-Shannon divergence, D KL is the Kullback-Leibler divergence; The feature determination module is used to determine the high-discrimination feature set of the brain functional network generated by the VAE generative network of different dimensional latent spaces, and use the adaptive enhancement method adaboost to train the feature sets of different dimensional models to obtain the support vector machine (SVM); The obstacle recognition module is used to input the test set into the strong classifier to obtain the classification result of mild cognitive impairment (MCI) to determine the user's cognitive impairment recognition result.

7. An electronic device, characterized in that: It includes: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method for constructing a functional connection network according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for constructing a functional connection network described in any one of claims 1 to 5 is implemented.

Images