Brain image classification method, device, and electronic equipment based on hierarchical graph convolution.
By optimizing brain image classification through hierarchical graph convolutional networks and edge weight encoding mechanisms, the problem of insufficient generalization in existing brain image classification models is solved, achieving higher feature accuracy and generalization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUZHOU INST OF BIOMEDICAL ENG & TECH CHINESE ACADEMY OF SCI
- Filing Date
- 2023-05-24
- Publication Date
- 2026-04-21
AI Technical Summary
Existing graph convolution-based brain image classification methods do not fully utilize hierarchical network structures and prior knowledge of brain regions, resulting in insufficient generalization performance of classification models across populations of different ages, genders, and image acquisition devices.
A hierarchical graph convolutional network structure is adopted. By sparsifying brain network features and using prior brain region knowledge to filter features, non-image information is fused by combining edge weight encoding mechanism to optimize graph embedding representation.
It improves the feature representation accuracy and generalization ability of brain image classification models, and reduces the differences in classification performance among different populations and devices.
Smart Images
Figure CN116721281B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, and in particular to a brain image classification method, apparatus, and electronic device based on hierarchical graph convolution. Background Technology
[0002] Unlike common convolutional neural networks (CNNs), graph convolutional networks (GCNNs) enable "convolutional" learning on graph-structured data. In recent years, focusing on brain network features, GCNNs have demonstrated good performance in brain image classification tasks. However, brain network features constructed using GCNNs are high-dimensional, which can easily lead to overfitting in classification models. Without feature optimization techniques for selection, the accuracy of feature representation may be affected. Furthermore, current GCNN classification models do not consider non-image information such as age, gender, and image acquisition device, which can impact the generalization performance of the classification model across different image acquisition devices and populations.
[0003] Most existing graph convolution-based brain image classification methods do not use hierarchical graph convolution network structures, but only include individual graph convolution networks. They use threshold sparsity methods to optimize brain network features, do not make full use of prior knowledge of brain regions, and the constructed graph convolution models use brain regions as nodes, which affects the generalization of classification models among people of different ages, genders, and image acquisition devices.
[0004] Therefore, there is an urgent need for a brain image classification method based on hierarchical graph convolution to solve the above problems. Summary of the Invention
[0005] To address the aforementioned problems, the first objective of this invention is to provide a brain image classification method based on hierarchical graph convolution, comprising the following steps:
[0006] Obtain resting-state functional magnetic resonance images of the training set and construct functional brain network features of the training set;
[0007] The training set functional brain network features are sparsified to obtain sparsed functional brain network features of the training set. Meaningful training set functional brain network features are selected as inputs for individual graph convolutional networks.
[0008] The sparse functional brain network features of the training set are input into the individual graph convolutional network model for training, and the graph embedding representation of the sparse functional brain network features of the training set is obtained. The optimal individual graph convolutional network model is saved and denoted as the first layer individual graph convolutional network model.
[0009] A group graph convolutional network model is constructed based on the graph embedding representation, and the edge weight parameters of the group graph convolutional network model are determined by the edge weight encoding mechanism. The optimal group graph convolutional network model is saved and denoted as the second-layer group graph convolutional network model.
[0010] We acquired resting-state functional magnetic resonance images of the test set, constructed functional brain network features of the test set, and derived sparse functional brain network features of the test set.
[0011] The sparse functional brain network features of the test set are input into the first layer of the individual graph convolutional network model to obtain the graph embedding representation of the test set.
[0012] The test set graph embedding representation is input into the second-layer group graph convolutional network model, and the classification results of the resting-state functional magnetic resonance images of the test set are obtained statistically.
[0013] More preferably, the input to the hierarchical graph convolutional network is the sparse brain network feature input training set individual graph convolutional network.
[0014] Preferably, the construction of the functional brain network features of the training set includes: calculating the average time series of 116 brain regions in the resting-state functional magnetic resonance imaging based on the AAL template, calculating the correlation coefficient between paired ROIs of the average time series using Pearson correlation, which is used to characterize the functional connectivity between pairs of brain regions, and constructing the functional brain network features of the resting-state functional magnetic resonance imaging.
[0015] Preferably, the sparse functional brain network features specifically include: performing a sparsification operation on the functional brain network features based on prior brain region knowledge, and selecting some brain network features as input to a hierarchical graph convolutional network.
[0016] Preferably, the graph embedding representation specifically includes: dividing the training set into multiple subsets of the training set according to the batch size, and inputting each subset into an individual graph convolutional network for training to obtain a graph embedding representation of each training set subset sample.
[0017] Preferably, the edge weight encoding is based on the edge weight encoding mechanism to encode the similarity of non-image information, and the similarity encoding of non-image information is multiplied with the original image kernel similarity result.
[0018] Preferably, the edge weight parameters in the group graph convolutional network are configured using the edge weight encoding mechanism, the group graph convolutional network is initialized, and the optimal model is retained after iterative training.
[0019] The second objective of this invention is to provide a brain image classification device based on hierarchical graph convolution, comprising:
[0020] Image acquisition unit, used to acquire resting-state functional magnetic resonance images;
[0021] Sparsification units, based on prior knowledge of brain regions, optimize brain network features and are used to sparsify functional brain network features.
[0022] The first layer of individual graph convolutional network model units is used to output graph embedding representations of brain network features;
[0023] The second-layer group graph convolutional network model unit is used to determine the edge weight parameters of the group graph convolutional network and initialize the group graph convolutional network.
[0024] Preferably, the sparsification module performs sparsification on the functional brain network features based on prior brain region knowledge, and selects some brain network features as input to the hierarchical graph convolutional network.
[0025] Preferably, the edge weight encoding module encodes non-image information based on the edge weight encoding mechanism, multiplies it with the graph kernel similarity result, and initializes the group graph convolutional network.
[0026] The third objective of this invention is to provide a computer-readable storage medium storing program instructions thereon, which, when executed, implement the brain image classification method based on hierarchical graph convolution as described above.
[0027] The fourth objective of this invention is to provide an electronic device comprising: a processor and a memory, wherein the memory is used to store one or more programs; when the one or more programs are executed by the processor, the brain image classification method based on hierarchical graph convolution as described above is implemented.
[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0029] This invention provides a brain image classification method based on hierarchical graph convolution. This method utilizes prior brain region sparsification to achieve sparsification of brain network features. Based on prior brain regions, statistical analysis is performed to screen out brain network features with significant differences, i.e., features with statistical significance, thus optimizing high-dimensional brain network features. Simultaneously, an edge weight encoding mechanism is used to optimize the edge representation of the group graph network, incorporating non-image information on top of the original graph kernel similarity, thereby improving the generalization ability of the classification model.
[0030] This invention provides a brain network sparsification unit based on prior brain regions, which optimizes brain network features and filters brain network features based on prior brain region knowledge, thereby improving the accuracy of feature representation. Secondly, it also proposes an edge weight encoding unit that integrates image features and non-image information to optimize the modeling of the correlation between nodes, so that the classification model has better generalization ability. Attached Figure Description
[0031] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and constitute an undue limitation thereof. In the drawings...
[0032] Figure 1 A flowchart of a brain image classification method based on hierarchical graph convolution provided in this embodiment of the invention;
[0033] Figure 2 A block diagram of a brain image classification device based on hierarchical graph convolution provided in an embodiment of the present invention;
[0034] Figure 3 A schematic diagram illustrating the specific process of a brain image classification method based on hierarchical graph convolution provided in an embodiment of the present invention;
[0035] Figure 4 A schematic diagram of a computer-readable storage medium provided in an embodiment of the present invention;
[0036] Figure 5 A schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0037] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.
[0038] Example 1
[0039] This invention provides a brain image classification method based on hierarchical graph convolution, such as... Figure 1 As shown, the process includes acquiring images, training the network model, testing the network model, and statistically analyzing the classification results, specifically including the following steps:
[0040] S1. Obtain resting-state functional magnetic resonance images of the training set and construct functional brain network features of the training set;
[0041] S2. Sparsify the training set functional brain network features to obtain the training set sparsed functional brain network features, and select meaningful training set functional brain network features as input to the individual graph convolutional network.
[0042] S3. Input the sparse functional brain network features of the training set into the individual graph convolutional network model for training, obtain the graph embedding representation of the sparse functional brain network features of the training set, and save the optimal individual graph convolutional network model, which is denoted as the first layer individual graph convolutional network model.
[0043] S4. Construct a group graph convolutional network model based on the graph embedding representation, and use the edge weight encoding mechanism to determine the edge weight parameters of the group graph convolutional network model, and save the optimal group graph convolutional network model, which is denoted as the second-layer group graph convolutional network model.
[0044] S5. Obtain resting-state functional magnetic resonance images of the test set and construct functional brain network features of the test set.
[0045] S6. Sparsify the functional brain network features of the test set;
[0046] S7. Input the sparse functional brain network features of the test set into the trained first-layer individual graph convolutional network model to obtain the graph embedding representation of the test set samples;
[0047] S8. Input the graph embedding representation of the test set samples into the second-layer population graph convolutional network model, and statistically obtain the classification results of the resting-state functional magnetic resonance images of the test set.
[0048] More preferably, the input to the hierarchical graph convolutional network is the sparse brain network feature input training set individual graph convolutional network.
[0049] Preferably, the construction of the functional brain network features of the training set includes: calculating the average time series of 116 brain regions in the resting-state functional magnetic resonance imaging based on the AAL template, calculating the correlation coefficient between paired ROIs of the average time series using Pearson correlation, which is used to characterize the functional connectivity between pairs of brain regions, and constructing the functional brain network features of the resting-state functional magnetic resonance imaging.
[0050] More preferably, the AAL template defines 116 brain regions, and the Pearson correlation coefficient is used to determine the correlation between pairs of brain regions to construct functional brain network features.
[0051] Preferably, the sparse functional brain network features specifically include: performing a sparsification operation on the functional brain network features based on prior brain region knowledge, and selecting some brain network features as input to a hierarchical graph convolutional network.
[0052] Preferably, the graph embedding representation specifically includes: dividing the training set into multiple subsets of the training set according to the batch size, and inputting each subset into an individual graph convolutional network for training to obtain a graph embedding representation of each training set subset sample.
[0053] Preferably, the edge weight encoding is based on the edge weight encoding mechanism to encode the similarity of non-image information, and the similarity encoding of non-image information is multiplied with the original image kernel similarity result.
[0054] Preferably, the edge weight parameters in the group graph convolutional network are configured using the edge weight encoding mechanism, the group graph convolutional network is initialized, and the optimal model is retained after iterative training.
[0055] Preferably, the sparse functional brain network features specifically include: performing a sparsification operation on the functional brain network features based on prior brain region knowledge, and selecting a portion of the brain network features as input to a hierarchical graph convolutional network, wherein the prior brain region knowledge is relevant brain regions analyzed from different brain diseases, and this is used as prior brain region knowledge.
[0056] Preferably, the graph embedding representation specifically includes: dividing the training set into multiple subsets of the training set according to the batch size, and inputting each subset into an individual graph convolutional network for training to obtain a graph embedding representation of each training set subset sample.
[0057] Preferably, the edge weight encoding is based on the edge weight encoding mechanism to encode the similarity of non-image information, and the similarity encoding of non-image information is multiplied with the original image kernel similarity result.
[0058] Preferably, the edge weight parameters in the group graph convolutional network are configured using the edge weight encoding mechanism, the group graph convolutional network is initialized, and the optimal model is retained after iterative training.
[0059] A further preferred approach is to preprocess the resting-state functional magnetic resonance imaging (rs-fMRI) images of each subject in the training set during the training phase, calculating the average time series for each subject across 116 brain regions based on the AAL116 template. Then, Pearson correlation is used to calculate the correlation coefficients between pairs of brain regions, characterizing the functional connectivity between these regions and thus constructing functional brain network features. Next, the functional brain network undergoes sparsification based on prior brain region knowledge to select meaningful brain network features, which are then used as input to a hierarchical graph convolutional network. The training set is divided into multiple training subsets according to batch size, and each subset is then fed into the first layer of the individual graph convolutional network for training, outputting a graph embedding representation for each training sample. The performance of the current model is evaluated using a validation set, and the optimal model parameters are retained after iterative training. The second-layer group graph convolutional network model is constructed based on the graph embedding representation of the individual graph convolutional network. The edge weights in the group graph network are designed using the proposed edge weight encoding mechanism to initialize the group graph network. Similarly, after iterative training, the classification results of all samples in the training set are obtained, and the optimal model is retained.
[0060] A further preferred approach is to preprocess the rs-fMRI images of each subject in the test set during the testing phase to construct a functional brain network, and then perform prior brain region sparsification to obtain a sparsified functional brain network. The sparsified brain network of the test set is then input into the optimal model saved during the training phase of the first-layer individual graph convolutional network to obtain the graph embedding representation of the test set samples. Next, an edge weight encoding mechanism is used to initialize the population graph network, and the graph embedding representation of the test set is input into the initialized second-layer population graph convolutional network model. The classification results of the test set are then statistically analyzed.
[0061] This invention provides a brain image classification method based on hierarchical graph convolution. This method utilizes prior brain region sparsification to achieve sparsification of brain network features. Based on statistical analysis of prior brain regions, it filters out brain network features with significant differences, thereby optimizing high-dimensional brain network features. Simultaneously, it employs an edge weight encoding mechanism to optimize the edge representation of the group graph network, incorporating non-image information on top of the original graph kernel similarity, thus improving the generalization ability of the classification model.
[0062] This invention provides a brain network sparsification unit based on prior brain regions, which optimizes brain network features and filters brain network features based on prior brain region knowledge to improve the accuracy of feature representation. Secondly, it also proposes an edge weight encoding unit that integrates image features and non-image information to reduce the impact of non-image information such as different subject ages, genders, and image acquisition devices on classification performance, thereby enabling the classification model to have better generalization.
[0063] Example 2
[0064] This invention provides a brain image classification device 100 based on hierarchical graph convolution, comprising:
[0065] Image acquisition unit 101 is used to acquire resting-state functional magnetic resonance images;
[0066] The sparsification unit 102, based on prior knowledge of brain regions, enables the optimized expression of brain network features and is used to sparsify functional brain network features.
[0067] The first layer of individual graph convolutional network model unit 103 is used to output graph embedding representations of brain network features;
[0068] The second-layer group graph convolutional network model unit 104 is used to determine the edge weight parameters of the group graph convolutional network, initialize the group graph convolutional network, and encode the edge weights of the group graph convolutional network model using an edge weight encoding mechanism.
[0069] Preferably, considering the high dimensionality of brain network features, the sparsification module performs sparsification operations on the functional brain network features based on prior brain region knowledge and statistical analysis results, selecting brain network features with significant differences from the high-dimensional brain network features, which has good accuracy.
[0070] Preferably, the edge weight encoding module encodes non-image information based on the edge weight encoding mechanism, multiplies it with the graph kernel similarity result, and initializes the group graph convolutional network.
[0071] Further preferably, the embodiments of the present invention provide an edge weight encoding module that integrates image features and non-image information, reducing the impact of non-image information such as different subject ages, genders, and image acquisition devices on classification performance, providing more information for the training and learning of graph convolutional networks, thereby improving the performance of the classification model and enhancing its generalization ability.
[0072] Further preferably, the embodiments of the present invention provide performance verification experimental results of the prior brain region sparsification module and the edge weight encoding module, as shown in Table 1.
[0073]
[0074] Table 1
[0075] The table above shows the experimental results of the hierarchical graph convolutional network, the prior brain region sparse graph convolutional network, and the edge weighted encoding graph convolutional network for accuracy, precision, area under the receiver operating characteristic curve (AUC), recall, and mean (F1-score).
[0076] This invention provides a brain image classification device based on hierarchical graph convolution. The device includes a first-layer individual graph convolutional network model unit based on prior brain regions, which optimizes brain network features by selecting features based on prior brain region knowledge, thereby improving the accuracy of feature representation. Secondly, a second-layer group graph convolutional network model unit integrates image features and non-image information. This unit utilizes an edge weight encoding mechanism to encode the edge weights of the group graph convolutional network model, reducing the impact of non-image information such as different subject ages, genders, and image acquisition devices on classification performance, thus enabling the classification model to have better generalization ability.
[0077] Example 3
[0078] A computer-readable storage medium, such as Figure 4 As shown, it stores program instructions, and when the program instructions are executed, they implement the brain image classification method based on hierarchical graph convolution as described in Example 1.
[0079] Example 4
[0080] An electronic device 200, such as Figure 5 As shown, it includes: a memory 201 and a processor 202, the memory being used to store one or more programs; when the one or more programs are executed by the processor, the brain image classification method based on hierarchical graph convolution as described in Embodiment 1 is implemented.
[0081] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented in software or in combination with necessary hardware. The technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several computer program instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described above according to the embodiments of this application.
[0082] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.
[0083] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be used in various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.
[0084] The apparatus, electronic device, computer storage medium and method provided in the embodiments of this specification are corresponding to each other. Therefore, the apparatus, electronic device and computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device and computer storage medium will not be described again here.
[0085] The apparatus or unit described in the above embodiments can be implemented by a computer chip or physical entity, or by a product with a certain function. For ease of description, the above apparatus is described by dividing it into various units according to their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.
[0086] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, apparatus, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0087] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0090] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0091] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. Especially for system embodiments, since they are fundamentally similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0092] The above are merely embodiments of this specification and are not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.
Claims
1. A brain image classification method based on hierarchical graph convolution, characterized in that, Includes the following steps: Obtain resting-state functional magnetic resonance images of the training set and construct functional brain network features of the training set; The training set functional brain network features are sparsified to obtain sparsed functional brain network features of the training set. Meaningful training set functional brain network features are selected as inputs for individual graph convolutional networks. The sparse functional brain network features of the training set are input into the individual graph convolutional network model for training, and the graph embedding representation of the sparse functional brain network features of the training set is obtained. The optimal individual graph convolutional network model is saved and denoted as the first layer individual graph convolutional network model. A group graph convolutional network model is constructed based on the graph embedding representation, and the edge weight parameters of the group graph convolutional network model are determined by the edge weight encoding mechanism. The optimal group graph convolutional network model is saved and denoted as the second-layer group graph convolutional network model. We acquired resting-state functional magnetic resonance images of the test set, constructed functional brain network features of the test set, and derived sparse functional brain network features of the test set. The sparse functional brain network features specifically include: based on prior brain region knowledge, performing sparsification on the functional brain network features, and selecting some brain network features as input to a hierarchical graph convolutional network; wherein the prior brain region knowledge is relevant brain regions analyzed from different brain diseases. The sparse functional brain network features of the test set are input into the first layer of the individual graph convolutional network model to obtain the graph embedding representation of the test set. The test set graph embedding representation is input into the second-layer group graph convolutional network model, and the classification results of the resting-state functional magnetic resonance images of the test set are obtained statistically.
2. The brain image classification method based on hierarchical graph convolution as described in claim 1, characterized in that, The construction of functional brain network features for the training set includes: calculating the average time series of 116 brain regions in resting-state functional magnetic resonance images based on the AAL template, calculating the correlation coefficient between paired ROIs of the average time series using Pearson correlation, which is used to characterize the functional connectivity between pairs of brain regions, and constructing functional brain network features of resting-state functional magnetic resonance images.
3. The brain image classification method based on hierarchical graph convolution as described in claim 1, characterized in that, The graph embedding representation specifically includes: dividing the training set into multiple subsets according to the batch size, and inputting each subset into an individual graph convolutional network for training to obtain the graph embedding representation of each training set subset sample.
4. The brain image classification method based on hierarchical graph convolution as described in claim 1, characterized in that, The edge weight encoding is based on the edge weight encoding mechanism to encode the similarity of non-image information, and then multiplies the similarity encoding of non-image information with the original image kernel similarity result.
5. The brain image classification method based on hierarchical graph convolution as described in claim 1, characterized in that, The edge weight parameters in the group graph convolutional network are configured using the edge weight encoding mechanism, the group graph convolutional network is initialized, and the optimal model is retained after iterative training.
6. A brain image classification device based on hierarchical graph convolution, used to perform the brain image classification method based on hierarchical graph convolution as described in any one of claims 1-5, characterized in that, include: Image acquisition unit, used to acquire resting-state functional magnetic resonance images; The sparsification unit, based on prior brain region knowledge, optimizes the representation of brain network features, sparsifies functional brain network features, and selects some brain network features as input to the hierarchical graph convolutional network; wherein the prior brain region knowledge is relevant brain regions analyzed from different brain diseases. The first layer of individual graph convolutional network model units is used to output graph embedding representations of brain network features; The second-layer group graph convolutional network model unit is used to determine the edge weight parameters of the group graph convolutional network and initialize the group graph convolutional network.
7. The brain image classification device based on hierarchical graph convolution as described in claim 6, characterized in that, The edge weight encoding module encodes non-image information based on the edge weight encoding mechanism, multiplies it with the graph kernel similarity result, and initializes the group graph convolutional network.
8. A computer-readable storage medium, characterized in that, It stores program instructions, and when the program instructions are executed, they implement the brain image classification method based on hierarchical graph convolution as described in any one of claims 1-5.
9. An electronic device, characterized in that, include: Processor and memory, the memory being used to store one or more programs; When the processor executes the one or more programs, it implements the brain image classification method based on hierarchical graph convolution as described in any one of claims 1-5.