A brain network classification method, system, electronic device and medium

Abstract

The application discloses a brain network classification method and system, an electronic device and a medium, and relates to the field of brain network classification. The method comprises the following steps: acquiring resting-state functional magnetic resonance imaging data of a person to be classified; preprocessing the resting-state functional magnetic resonance imaging data to obtain processed resting-state functional magnetic resonance imaging data; determining a Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain a functional brain network of the person to be classified; and determining a brain network category of the person to be classified by using a brain network classification model according to the functional brain network. The brain network classification model is obtained by training a graph autoencoder classification model by using a training data set. The graph autoencoder classification model comprises a graph autoencoder for embedding a cross-group high-order brain network, a decoder and a fully connected layer classifier. The cross-group high-order brain network is determined according to a functional brain network of a subject and a structural brain network of the subject. The application improves the accuracy of brain network classification.

Description

Technical Field

[0001] This invention relates to the field of brain network classification, and in particular to a brain network classification method, system, electronic device, and medium. Background Technology

[0002] Most brain diseases involve widespread structural and functional abnormalities. Exploring the complex network topological changes caused by brain diseases has unique advantages in revealing the characteristics of neuropathological diseases, and constructing brain networks is an effective means of studying the working patterns between different brain regions. Previous studies have shown significant differences in the topological properties of brain networks between patients and healthy individuals, but revealing the underlying pathological mechanisms of the degradation of the brain connectome from the healthy group to patients with brain diseases remains a challenge. Currently, brain network analysis methods are widely used in the study of brain diseases. Regarding population differences, existing studies mainly examine the differences in complex interactions between two groups of individuals. However, existing methods mainly focus on low-order correlations between paired brain regions, neglecting the complex high-order information between multiple brain regions.

[0003] Brain activity is a highly complex process. Each brain region interacts with many other brain regions, and this interaction is not a simple one-to-one relationship but a complex many-to-many relationship, thus forming a vast brain network in which all nodes are closely intertwined. Low-order correlations refer to one-to-one relationships. For example, in brain networks constructed using the currently mainstream Pearson correlation coefficient, each element represents the correlation between any two brain regions, without considering the relationship between a particular brain region and several other brain regions. Therefore, it ignores the higher-order information of the many-to-many relationships, leading to inaccurate brain network classification. Summary of the Invention

[0004] The purpose of this invention is to provide a brain network classification method, system, electronic device, and medium to improve the accuracy of brain network classification.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A brain network classification method, comprising:

[0007] Acquire resting-state functional magnetic resonance imaging data of the subjects to be classified;

[0008] The resting-state functional magnetic resonance imaging (fMRI) data is preprocessed to obtain processed resting-state fMRI data; the preprocessing includes correction, alignment, normalization, and brain region segmentation.

[0009] The Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data was determined to obtain the functional brain network of the subject to be classified.

[0010] Based on the functional brain network, a brain network classification model is used to determine the brain network category of the subject to be classified; wherein, the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder, a decoder, and a fully connected layer classifier embedded with a cross-group higher-order brain network; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network.

[0011] Optionally, the resting-state functional magnetic resonance imaging (fMRI) data is preprocessed to obtain processed resting-state fMRI data, specifically including:

[0012] The resting-state functional magnetic resonance imaging data were corrected, aligned, and normalized to the EPI template using the SPM8 toolbox of DPARSF2.0 to obtain normalized resting-state functional magnetic resonance imaging data.

[0013] The normalized resting-state functional magnetic resonance imaging (fMRI) data were divided into brain regions using the AAL template. For each brain region, the average resting-state fMRI time series of all voxels was used as the time series of that brain region to obtain the processed resting-state fMRI data.

[0014] Optionally, the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data is determined to obtain the functional brain network of the subject to be classified, specifically including:

[0015] Calculate the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data;

[0016] The functional brain network of the subject to be classified is determined based on the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data.

[0017] Optionally, the graph autoencoder classification model is trained using a training dataset, specifically including:

[0018] Build the training dataset;

[0019] Based on the training dataset, the graph autoencoder classification model is trained using the 10-fold cross-validation method to obtain a brain network classification model.

[0020] Optionally, a training dataset is constructed, specifically including:

[0021] Acquire resting-state functional magnetic resonance imaging (fMRI) data and diffusion magnetic resonance imaging (DMRI) data of the subjects;

[0022] The resting-state functional magnetic resonance imaging (fMRI) data of the subject were preprocessed to obtain the processed resting-state fMRI data of the subject.

[0023] The PANDA kit was used to correct the distortion of the subject's diffusion magnetic resonance imaging (DMRI) data. The TrackVis was used to obtain the fiber images of the corrected DMRI data. The AAL template was used to divide the brain regions of the fiber images of the DMRI data to obtain the subject's processed DMRI data.

[0024] The functional brain network of the subject was calculated based on the processed resting-state functional magnetic resonance imaging data of the subject.

[0025] Based on the processed diffusion magnetic resonance imaging data of the subject, the structural brain network of the subject was calculated;

[0026] A training dataset is constructed based on the subject's functional brain network, the subject's structural brain network, and the corresponding category labels.

[0027] Optionally, cross-group higher-order brain networks are determined based on the subjects' functional brain networks and structural brain networks, specifically including:

[0028] The participants were divided into healthy and sick groups;

[0029] Based on the functional brain networks of the healthy group and the diseased group, calculate the average functional brain network of the healthy group and the average functional brain network of the diseased group.

[0030] Calculate the average structural brain network of the healthy group based on the structural brain network of the healthy group;

[0031] Using the PageRank algorithm, the brain region centrality distribution of the average functional brain network of the healthy group and the brain region centrality distribution of the average functional brain network of the diseased group were determined.

[0032] The cross-group higher-order brain network is determined based on the central distribution of brain regions in the average functional brain network of the healthy group, the central distribution of brain regions in the average functional brain network of the diseased group, and the average structural brain network of the healthy group.

[0033] A brain network classification system, comprising:

[0034] The data acquisition module is used to acquire resting-state functional magnetic resonance imaging data of the subjects to be classified.

[0035] The preprocessing module is used to preprocess the resting-state functional magnetic resonance imaging (fMRI) data to obtain processed resting-state fMRI data; the preprocessing includes correction, alignment, normalization, and brain region segmentation;

[0036] A functional brain network determination module is used to determine the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain the functional brain network of the subject to be classified.

[0037] A classification module is used to determine the brain network category of the subject based on the functional brain network using a brain network classification model; wherein the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network.

[0038] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the brain network classification method described above.

[0039] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described brain network classification method.

[0040] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0041] This invention provides a brain network classification method, system, electronic device, and medium. The method involves acquiring resting-state functional magnetic resonance imaging (fMRI) data of the subject to be classified; preprocessing the resting-state fMRI data to obtain processed resting-state fMRI data; determining the Pearson correlation coefficient of the processed resting-state fMRI data to obtain the functional brain network of the subject; and using a brain network classification model based on the functional brain network to determine the brain network category of the subject. The brain network classification model is obtained by training a graph autoencoder classification model using a training dataset. The graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier. The cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network. This invention improves the accuracy of brain network classification by embedding a cross-group higher-order brain network into the graph autoencoder, considering the relationship between a certain brain region and several other brain regions. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A flowchart of the brain network classification method provided by this invention;

[0044] Figure 2 This is a flowchart illustrating the construction and training process of the graph autoencoder classification model provided by the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The purpose of this invention is to provide a brain network classification method, system, electronic device, and medium to improve the accuracy of brain network classification.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Example 1

[0049] like Figure 1 As shown, the present invention provides a brain network classification method, comprising:

[0050] Step 101: Acquire resting-state functional magnetic resonance imaging data of the subjects to be classified.

[0051] Step 102: Preprocess the resting-state functional magnetic resonance imaging data to obtain processed resting-state functional magnetic resonance imaging data; the preprocessing includes correction, alignment, normalization and brain region segmentation.

[0052] As an optional implementation, step 102 specifically includes:

[0053] The resting-state functional magnetic resonance imaging (fMRI) data was corrected, aligned, and normalized to the EPI template using the SPM8 toolbox of DPARSF2.0 to obtain normalized resting-state fMRI data.

[0054] The normalized resting-state functional magnetic resonance imaging (fMRI) data was divided into brain regions using the AAL (Anatomical Automatic Labeling) template. For each brain region, the average resting-state fMRI time series of all voxels was used as the time series of that brain region to obtain the processed resting-state fMRI data.

[0055] Step 103: Determine the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain the functional brain network of the subject to be classified.

[0056] As an optional implementation, step 103 specifically includes:

[0057] Calculate the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data.

[0058] The functional brain network of the subject to be classified is determined based on the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data.

[0059] Step 104: Based on the functional brain network, determine the brain network category of the subject using a brain network classification model; wherein, the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network. The brain network category is either healthy or diseased.

[0060] As an alternative implementation method, such as Figure 2 As shown, the graph autoencoder classification model is trained using the training dataset, specifically including:

[0061] S1: Construct the training dataset. The training dataset includes the subject's functional brain network + the subject's structural brain network + labels. The test dataset or the process of testing the model uses the subject's functional brain network.

[0062] In practical applications, S1 specifically includes:

[0063] S11: Acquire resting-state functional magnetic resonance imaging (fMRI) data and diffusion magnetic resonance imaging (DMRI) data of the subjects.

[0064] S12: Preprocess the resting-state functional magnetic resonance imaging (fMRI) data of the subject to obtain the processed resting-state fMRI data of the subject.

[0065] S13: The PANDA kit is used to correct the distortion of the subject's diffusion magnetic resonance imaging (DMRI) data. The TrackVis is used to obtain the fiber images of the corrected DMRI data. The AAL template is used to divide the brain regions of the fiber images of the DMRI data to obtain the subject's processed DMRI data.

[0066] In practical applications, resting-state functional magnetic resonance imaging (rs-fMRI) data and diffusion magnetic resonance imaging (DTI) data of subjects are acquired, and corresponding data preprocessing is performed.

[0067] Preprocessing of rs-fMRI: Using the SPM8 toolbox in DPARSF version 2.0, the rs-fMRI images were first corrected and aligned to account for head movement during scanning and normalized to the EPI template. Then, the AAL (Anatomical Automatic Labeling) template was used to divide the rs-fMRI images into 90 brain regions. For each brain region, the average rs-fMRI time series of all voxels was used as the time series for that region. The time series dimension of the rs-fMRI data used in this invention is 240. The final preprocessed rs-fMRI data (processed resting-state functional magnetic resonance imaging data of the subject) is a 90×240 matrix, representing the temporal signals of the 90 divided brain regions.

[0068] DTI preprocessing: Using the PANDA kit, distortion correction was performed on the DTI data, and fiber images were obtained using TrackVis. Anatomical regions were segmented according to the AAL template to obtain preprocessed DTI data (processed diffusion magnetic resonance imaging data of the subjects). Each preprocessed DTI data is a 90×90 element matrix, representing the number of physical fibers between pairs of brain regions.

[0069] S14: Calculate the functional brain network of the subject based on the processed resting-state functional magnetic resonance imaging data of the subject.

[0070] S15: Calculate the subject's structural brain network based on the subject's processed diffusion magnetic resonance imaging data.

[0071] In practical applications, functional and structural brain networks are further calculated from preprocessed rs-fMRI and preprocessed DTI data.

[0072] Computational functional brain networks were performed. For temporal signals from 90 brain regions, Pearson correlation coefficients were calculated for each pair of regions. The specific calculation method is as follows: Where x and y are the temporal signal vectors of two brain regions, Cov(x,y) is the covariance of x and y, and Var(x) and Var(y) are the variances of x and y, respectively. The resulting 90×90 matrix represents the functional brain network of each subject.

[0073] Calculate the structural brain network. The element matrix represents the number of fibers between brain regions, which can be seen as the strength between brain regions. This matrix is ​​the structural brain network of the subject.

[0074] S16: Construct a training dataset based on the subject's functional brain network, the subject's structural brain network, and the corresponding category labels.

[0075] S2: Based on the training dataset, the graph autoencoder classification model is trained using the ten-fold cross-validation method to obtain the brain network classification model.

[0076] In practical applications, the first step is to identify the higher-order brain networks across groups. These higher-order brain networks are determined based on the subjects' functional and structural brain networks, and specifically include:

[0077] The participants were divided into a healthy group and a disease group.

[0078] Based on the functional brain networks of the healthy group and the diseased group, the average functional brain network of the healthy group and the average functional brain network of the diseased group are calculated.

[0079] The average structural brain network of the healthy group is calculated based on the structural brain network of the healthy group.

[0080] In practical applications, the average functional brain network of the two groups and the structural functional brain network of the healthy group are calculated.

[0081] The healthy group and the diseased group were designated as Group A and Group B, respectively. The average functional brain network and average structural brain network of Group A were respectively... and Where, N A This refers to the number of subjects in group A. The functional brain network F of each subject in group A i The result of summing and averaging The structural brain network S of each subject in group A i The result of summing and averaging. The average functional brain network of group B is Where, N B This refers to the number of subjects in group B. The functional brain network F of each subject in group B i The result of summing and averaging.

[0082] Using the PageRank algorithm, the brain region centrality distribution of the average functional brain network of the healthy group and the brain region centrality distribution of the average functional brain network of the diseased group were determined.

[0083] In practical applications, the brain region importance of the average functional brain network of the two groups is calculated and regarded as the central distribution of each brain region in the two groups.

[0084] Specifically, this study explores the potential pathological mechanisms leading to disease degeneration at different stages, such as how brain networks evolve from a healthy state to a diseased state. We can assume that the brain activity characteristics of the healthy and diseased states represent the source and target states of the disease process, respectively, and view this process as an interaction between brain regions in these two states. Existing literature indicates that brain networks exhibit small-world characteristics, with the overall brain's interaction and connectivity primarily maintained by certain core nodes within the network. Here, the PageRank algorithm is used to extract key nodes in the brain network, and the importance of nodes in PageRank is used to measure the centrality of brain regions. The PageRank algorithm is widely used in the web domain, emphasizing that if a webpage x is linked to an important page y, then x becomes important accordingly. The interaction between multiple brain regions in a brain network is analogous to referencing webpages on the World Wide Web. In this invention, we define... Where p(u) represents the importance of brain node u, B(u) represents all brain nodes linked to u, and N... v Define the out-degree of brain node v. First, initialize the node importance vector as follows: Where m is the number of brain regions. Then, a transition matrix M is constructed based on the degree between nodes, where each element M... uv The definition is as follows: Assume degree uv It represents the degree from node u to node v, indicating the number of edges in the network from node u to node v. uv When = 1, When degree uv When = 0, M uv =0. Then update the node importance vector by p = M × p until ||p|| is satisfied. k -p k-1 Up to ||2<ζ, k≤300. In the model, the brain network of the healthy group... and disease group brain network The node importance vector is calculated as p A =[p1,p2,…p m ]∈R m×1 and p B =[p1,p2,…p m ]∈Rm×1 p A and p B The larger the element, the stronger the centrality of the brain node.

[0085] The cross-group higher-order brain network is determined based on the central distribution of brain regions in the average functional brain network of the healthy group, the central distribution of brain regions in the average functional brain network of the diseased group, and the average structural brain network of the healthy group.

[0086] In practical applications, cross-group higher-order brain networks are calculated based on the brain region centrality distribution vectors of the two groups' brain networks (the brain region centrality distribution of the average functional brain network of the healthy group and the brain region centrality distribution of the average functional brain network of the diseased group).

[0087] During brain information processing, healthy subjects may experience physiological abnormalities in their bioelectrical signals, leading to functional abnormalities and ultimately regressing into patients with brain diseases. This evolution can be viewed as a process of brain activity propagation, in which central nodes with stronger interaction capabilities and higher information transmission efficiency tend to transmit more information to multiple nodes. This invention considers the centrality distribution of Group A (healthy group) and Group B (patient group) as the initial and target states of the model, respectively. Considering the brain's economy, the brain will use as little "energy consumption" as possible to achieve more information interaction during activity. Structural brain networks are a physiological basis; the larger the element in a structural brain connection, the stronger the interaction between two brain regions and the lower the transmission probability. The average structural brain network of Group A is used as an example. The model is embedded to constrain the transmission of two states. The aim is to explore higher-order interactions between cross-group brain networks, where each brain region may interact with multiple brain regions. Therefore, a Kantorovich-type optimal transmission method is introduced to solve the node centrality transmission problem. This method explores how to approximate the target distribution with the source distribution through inter-point mass transmission with minimal cost, and the mass of each node in the source distribution can be split and transmitted to multiple target nodes, which is consistent with the cross-group higher-order brain networks that this invention aims to explore.

[0088] The definition of cross-group higher-order brain networks proposed in this invention is as follows:

[0089]

[0090]

[0091] in, for The normalized inversion matrix, R M×M Represents a two-dimensional matrix of dimension M*M; R M T represents a one-dimensional vector with dimension M*1; ijLet T1 be an element in matrix T; M T represents a one-dimensional vector obtained by summing the elements of each row in T; ii The elements on the diagonal of matrix T; T T 1 M This represents a one-dimensional vector obtained by summing the elements of each column in T, where T is the cross-group high-order brain network to be obtained. <·,·> F Let T be the Frobenius inner product of two matrices. Here, the Sinkhorn algorithm is used as the optimization method for the above problem. It is worth noting that this model is used to explore the process of node importance diffusion from group A to group B, so the resulting matrix T effectively reveals the group differences between different groups. Furthermore, unlike previous brain networks based on low-order associations of paired brain regions, T is a high-order brain network whose propagation mechanism shows the relationships between multiple nodes.

[0092] Then, a graph autoencoder classification model is constructed based on the graph attention mechanism, and a cross-group high-order brain network is embedded in it, specifically:

[0093] 1. An encoder is constructed based on the graph self-attention mechanism, and a cross-group high-order brain network is embedded in the encoder.

[0094] To encode the topological brain network F and its node features X, nodes can focus on the features of their neighbors by specifying different attention weights for those neighbors. Assume that the input and output node features of each layer are H = [h1, h2, ..., h...]. M ]∈R M×D and H' = [h'1, h'2, ..., h' M ]∈R M×D ', then the convolution operation is defined as:

[0095] h' i =σ(∑ j∈N(i) att ij Wh j ).

[0096] Where W∈R D×D' Let N(i) be a trainable weight matrix, and N(i) be all neighboring nodes of node i; σ is the convolution operation. Note the coefficient att. ij The correlation between brain region i and its first-order neighbor j was quantified:

[0097] att ij =softmax(a(a) T [Wh i ||Wh j ]))×f ij ×t ij .

[0098] Where a is the Leaky ReLU function, a∈R 2D' It is a weight vector of learnable parameters. When measuring the correlation between two brain regions, comparing the features of the two nodes and their connections is important, as it represents the differences in brain connectivity groups. Therefore, when updating node features by calculating the attention coefficient between nodes, feature information from local neighbors, edge weights of the brain network, and inter-group discriminative brain connections is simultaneously aggregated. ij Also multiplied by the FC weight f ij and cross-group brain network weights t ij After an encoder with two layers of GATs, the original FC embedding Z∈R is then applied. M×T' middle.

[0099] Z=GATconv(g(GATconv(F,X))).

[0100] Where g(·) is the ReLU activation function, F is the subject's functional brain network, and X is the subject's functional magnetic resonance imaging time-series signal.

[0101] Local neighbor features are 0-1 matrices of a subject's functional brain network input during training or testing, with dimensions of 90×240. Then, the correlation between nodes is calculated using an attention mechanism; the correlation between two points is the aforementioned att. ij In the formula, a(a T [Wh i ||Wh j ])),Too Figure 2 α in ab .

[0102] The brain network edge weights feature is a functional brain network of a specific subject input during training or testing, with dimensions of 90×240. The elements in the matrix are the aforementioned att. ij f in the formula ij ,Too Figure 2 f in ab .

[0103] The inter-group discriminative brain connectivity features were obtained from a T-matrix, a cross-group high-order brain network with dimensions of 90×240. The elements in the matrix are the aforementioned att matrices. ij t in the formula ij ,Too Figure 2 t in ab .

[0104] 2. Construct a decoder.

[0105] To reconstruct the structural brain network S, the decoder module of GAE calculates the probability that there is an edge between two brain regions in Z, defined as:

[0106]

[0107] 3. Construct an MLP classifier.

[0108] The graph embedding features Z are input into an MLP classifier, and the log-softmax activation function is used to predict the label for each graph. Finally, GAE is trained (training process) to minimize the target structural brain network S and the reconstructed functional brain network. The loss function consists of a CrossEntropy loss and an Nll loss: (The text abruptly ends here, so the translation stops as much as possible.)

[0109]

[0110] Where Y and Y' are the true and predicted labels for each brain network, respectively, and λ is the scaling parameter for adjusting the CrossEntropy loss and Nll loss; E S The expectation for structural brain networks.

[0111] Finally, the constructed graph autoencoder classification model is trained, and the training process is as follows:

[0112] Ten-fold cross-validation is used to divide the training dataset into training and testing sets. The model is trained and tested within each fold, and the average of the test results (such as accuracy) for each fold is taken as the final model performance metric.

[0113] For training within each fold:

[0114] Steps S1-S2 are performed to obtain the cross-group high-order brain network T of the training set. This matrix will be embedded into att... ij In the calculation.

[0115] The input data to the model consists of each subject's functional brain network and corresponding labels (healthy or diseased), and the output is the predicted label.

[0116] For the test within each fold:

[0117] The input data is the subject's functional brain network, and the output is the predicted label.

[0118] The functional and structural brain network data of the subjects are input into a trained graph autoencoder model to obtain binary classification results. Specifically, the dataset is divided into a training set and a test set. For the training set, it is divided into two groups of data, and a cross-group high-order brain network is constructed. This network data is then embedded into the encoder of the graph autoencoder classification model. When using the test set data, the functional and structural brain network data of each data point in the test set are input into the model to obtain binary classification results.

[0119] Example 2

[0120] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a brain network classification system is provided below, including:

[0121] The data acquisition module is used to acquire resting-state functional magnetic resonance imaging data of the subjects to be classified.

[0122] The preprocessing module is used to preprocess the resting-state functional magnetic resonance imaging data to obtain processed resting-state functional magnetic resonance imaging data; the preprocessing includes correction, alignment, normalization and brain region segmentation.

[0123] The functional brain network determination module is used to determine the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain the functional brain network of the subject to be classified.

[0124] A classification module is used to determine the brain network category of the subject based on the functional brain network using a brain network classification model; wherein the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network.

[0125] Example 3

[0126] The present invention provides an electronic device, including: a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the brain network classification method of Embodiment 1.

[0127] Example 4

[0128] The present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the brain network classification method of Embodiment 1.

[0129] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0130] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A brain network classification method, characterized in that, include: Acquire resting-state functional magnetic resonance imaging data of the subjects to be classified; The resting-state functional magnetic resonance imaging (fMRI) data is preprocessed to obtain processed resting-state fMRI data; the preprocessing includes correction, alignment, normalization, and brain region segmentation. The Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data was determined to obtain the functional brain network of the subject to be classified. Based on the functional brain network, a brain network classification model is used to determine the brain network category of the subject to be classified; wherein, the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network; Cross-group higher-order brain networks were identified based on the subjects' functional and structural brain networks, specifically including: The participants were divided into healthy and sick groups; Based on the functional brain networks of the healthy group and the diseased group, calculate the average functional brain network of the healthy group and the average functional brain network of the diseased group. Calculate the average structural brain network of the healthy group based on the structural brain network of the healthy group; Using the PageRank algorithm, the brain region centrality distribution of the average functional brain network of the healthy group and the brain region centrality distribution of the average functional brain network of the diseased group were determined. The cross-group higher-order brain network is determined based on the central distribution of brain regions in the average functional brain network of the healthy group, the central distribution of brain regions in the average functional brain network of the diseased group, and the average structural brain network of the healthy group.

2. The brain network classification method according to claim 1, characterized in that, The resting-state functional magnetic resonance imaging (fMRI) data is preprocessed to obtain processed resting-state fMRI data, specifically including: The resting-state functional magnetic resonance imaging data were corrected, aligned, and normalized to the EPI template using the SPM8 toolbox of DPARSF2.0 to obtain normalized resting-state functional magnetic resonance imaging data. The normalized resting-state functional magnetic resonance imaging (fMRI) data were divided into brain regions using the AAL template. For each brain region, the average resting-state fMRI time series of all voxels was used as the time series of that brain region to obtain the processed resting-state fMRI data.

3. The brain network classification method according to claim 1, characterized in that, Determining the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain the functional brain network of the subject to be classified, specifically includes: Calculate the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data; The functional brain network of the subject to be classified is determined based on the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data.

4. The brain network classification method according to claim 1, characterized in that, The graph autoencoder classification model is trained using a training dataset, specifically including: Build the training dataset; Based on the training dataset, the graph autoencoder classification model is trained using the 10-fold cross-validation method to obtain a brain network classification model.

5. The brain network classification method according to claim 4, characterized in that, Constructing the training dataset specifically includes: Acquire resting-state functional magnetic resonance imaging (fMRI) data and diffusion magnetic resonance imaging (DMRI) data of the subjects; The resting-state functional magnetic resonance imaging (fMRI) data of the subject were preprocessed to obtain the processed resting-state fMRI data of the subject. The PANDA kit was used to correct the distortion of the subject's diffusion magnetic resonance imaging (DMRI) data. The TrackVis was used to obtain the fiber images of the corrected DMRI data. The AAL template was used to divide the brain regions of the fiber images of the DMRI data to obtain the subject's processed DMRI data. The functional brain network of the subject was calculated based on the processed resting-state functional magnetic resonance imaging data of the subject. Based on the processed diffusion magnetic resonance imaging data of the subject, the structural brain network of the subject was calculated; A training dataset is constructed based on the subject's functional brain network, the subject's structural brain network, and the corresponding category labels.

6. A brain network classification system, characterized in that, include: The data acquisition module is used to acquire resting-state functional magnetic resonance imaging data of the subjects to be classified. The preprocessing module is used to preprocess the resting-state functional magnetic resonance imaging (fMRI) data to obtain processed resting-state fMRI data; the preprocessing includes correction, alignment, normalization, and brain region segmentation; A functional brain network determination module is used to determine the Pearson correlation coefficient of the processed resting-state functional magnetic resonance imaging data to obtain the functional brain network of the subject to be classified. A classification module is used to determine the brain network category of the subject based on the functional brain network using a brain network classification model; wherein, the brain network classification model is obtained by training a graph autoencoder classification model using a training dataset; the training dataset includes the subject's functional brain network and corresponding category labels; the graph autoencoder classification model includes a graph autoencoder embedded with a cross-group higher-order brain network, a decoder, and a fully connected layer classifier; the cross-group higher-order brain network is determined based on the subject's functional brain network and the subject's structural brain network; Cross-group higher-order brain networks were identified based on the subjects' functional and structural brain networks, specifically including: The participants were divided into healthy and sick groups; Based on the functional brain networks of the healthy group and the diseased group, calculate the average functional brain network of the healthy group and the average functional brain network of the diseased group. Calculate the average structural brain network of the healthy group based on the structural brain network of the healthy group; Using the PageRank algorithm, the brain region centrality distribution of the average functional brain network of the healthy group and the brain region centrality distribution of the average functional brain network of the diseased group were determined. The cross-group higher-order brain network is determined based on the central distribution of brain regions in the average functional brain network of the healthy group, the central distribution of brain regions in the average functional brain network of the diseased group, and the average structural brain network of the healthy group.

7. An electronic device, characterized in that, include: A memory and a processor, the memory being used to store a computer program, the processor running the computer program to cause the electronic device to perform the brain network classification method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the brain network classification method according to any one of claims 1-5.

Images