Heterogeneity perception spectrogram convolutional network-based brain disease classification method and system

Abstract

The invention relates to the field of medical image processing, and provides a brain disease diagnosis classification method and system based on a heterogeneity perception spectrogram convolutional network. The method comprises the following steps: based on a resting state brain function image of a subject, extracting brain function connections, calculating Pearson correlation between the brain function connections, and obtaining node features of a crowd graph; according to the relation between different subjects, a phenotypic encoder is adopted to construct the weight of edges between nodes so as to construct a crowd graph; based on the crowd graph, calculating local similarity between the nodes and neighbor nodes thereof to obtain local similarity weights of all the nodes; based on the local similarity weights of all the nodes, using an MLP layer to obtain weights of different layers for guiding feature fusion among the layers to obtain node feature representations of all the layers; splicing the feature representations of all layers by adopting simplified spectrogram convolution to obtain multi-layer feature representations; and based on multilayer feature representation, obtaining a brain disease classification probability matrix, and obtaining a classification result.

Description

Technical Field

[0001] This invention relates to the field of medical image processing, and more particularly to a brain disease diagnosis classification and system based on heterogeneous perceptual spectral convolutional networks. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Brain diseases not only pose significant challenges to cognitive abilities but also severely impact patients' behavior, emotional regulation, and social interaction. The complexity of these diseases is reflected both in their diverse pathological mechanisms and the high heterogeneity of symptoms among different individuals, making accurate classification extremely difficult.

[0004] Machine learning and deep learning techniques are widely used in rs-fMRI data analysis to assist in the diagnosis of brain diseases. Currently, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and graph neural networks (GNNs) are commonly used deep learning models. The first two can extract spatial and temporal features from rs-fMRI data, achieving good results in individual-level diagnostic tasks. However, methods based on CNNs or RNNs primarily focus on individual individuals and struggle to characterize complex patterns at the group level. In contrast, models that incorporate inter-individual correlations are better able to learn group characteristics, thereby improving diagnostic accuracy and robustness. In fact, group diagnosis of brain diseases can be naturally modeled as a graph structure, where nodes represent individual imaging features (such as mild cognitive impairment or healthy controls), and edges characterize the similarities between individuals. GNNs have attracted widespread attention due to their effectiveness in handling complex relationships within graph structures. Most GNN-based brain disease prediction methods are based on individual-specific brain maps. In these maps, nodes typically represent specific brain regions, while edges represent relationships or connections between these regions. However, current GNNs primarily rely on imaging data, often neglecting key phenotypic data such as age, sex, and genotype. This limitation weakens the model's ability to capture comprehensive individual interactions.

[0005] To construct more robust population graphs and further improve the diagnostic performance of brain diseases, combining imaging data with phenotypic information is crucial. In recent years, researchers have proposed a series of graph neural network models based on population graphs to enhance the performance of brain disease diagnosis. These methods use individual image features as nodes in the population graph and utilize image similarity or clinical phenotypic similarity (such as age, sex, and scanning site) to construct edge connections, enabling the graph neural network model to integrate multiple individual-level information. While existing methods generally employ manually defined affinity functions (such as Euclidean distance and cosine similarity) to statically construct edges between population graph nodes, generating a fixed population graph structure, they neglect the nonlinear interactions between multimodal features. This results in the constructed edge weights failing to fully reflect the potential pathological mechanisms in the real population, limiting the expressive power of the graph structure and the model performance.

[0006] Furthermore, most existing methods are based on the homogeneity assumption of graph structures, which assumes that connected nodes belong to the same category. However, this assumption often does not hold true in brain disease diagnosis, especially in heterogeneous population graphs, where adjacent individuals may belong to healthy control groups and disease groups (such as Alzheimer's patients). This makes traditional graph neural networks unable to accurately capture heterogeneous neighbor information, reducing the model's ability to distinguish disease boundaries. Existing technologies generally ignore this heterogeneity problem, thus limiting the model's diagnostic performance and generalization ability. Summary of the Invention

[0007] To address the technical problems mentioned above, this invention provides a brain disease classification method and system based on heterogeneous perceptual graph convolutional networks. This invention employs multimodal data based on phenotypic encoders to dynamically learn complex relationships between individuals, and proposes a local similarity learning mechanism. By measuring the similarity between a node and its neighboring nodes through local topological information, the homogeneity level is evaluated, achieving accurate fusion of cross-layer features in the graph network model, thereby significantly improving the accuracy of brain disease classification.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a brain disease classification method based on heterogeneous perceptual spectral convolutional networks.

[0009] A brain disease classification method based on heterogeneous perceptual spectral convolutional networks includes: Based on the resting-state brain function images of the subjects, brain functional connectivity was extracted, and Pearson correlations between brain functional connectivity were calculated to obtain the node features of the population graph. According to the relationships between different subjects, a phenotypic encoder was used to construct the weights of the edges between nodes to construct the population graph. Based on the population graph, the local similarity between a node and its neighboring nodes is calculated to obtain the local similarity weights of all nodes. Based on the local similarity weights of all nodes, an MLP layer is used to obtain the weights of different layers, which are used to guide the feature fusion between layers to obtain the node feature representations of all layers. Simplified spectral graph convolution is used to concatenate the feature representations of all layers to obtain multi-layer feature representations. Based on the multi-layer feature representations, the brain disease classification probability matrix is ​​obtained, and the classification results are obtained.

[0010] Furthermore, based on the relationships between different subjects, a phenotypic encoder is used to construct the weights of the edges between nodes; expressed by the following formula:

[0011] in, Indicates individual subjects and individuals Edge weights between them and This is a multi-level mapping function, whose function is to map phenotypic features separately. and Mapped to the hidden feature space; It is a bilinear weight matrix. This indicates the bias term.

[0012] Furthermore, the method of calculating the local similarity between a node and its neighboring nodes based on the crowd graph to obtain the local similarity weight of all nodes includes: based on the crowd graph, introducing a nonlinear term and processing the nonlinear term through a multilayer perceptron to encode the distance variance between neighboring nodes to obtain the local similarity between a node and its neighboring nodes, and combining the local similarity of all nodes to obtain the local similarity weight of all nodes.

[0013] Furthermore, based on the local similarity weights of all nodes, an MLP layer is used to obtain weights for different layers, which are used to guide feature fusion between layers to obtain node feature representations for all layers; the method includes: inputting the local similarity weights of all nodes and their squares into an MLP layer to generate weights for different layers, and based on the weights of each layer and each S 2 The output of the GC module yields the node feature representations for each layer.

[0014] Furthermore, the simplified spectral graph convolution is used to concatenate the feature representations of all layers, expressed by the following formula:

[0015] in, Indicates the k-th S 2 The output of the GC module; L represents the number of times neighbor information is aggregated in the graph, i.e., the number of hops in the propagation process; It is the learned weight matrix; Indicates the process l The feature propagation matrix of the jump.

[0016] Furthermore, the feature propagation matrix is ​​represented by the following formula:

[0017]

[0018] in, Represents the adjacency matrix. Is with The corresponding degree matrix, It is the matrix after adding self-loops to the adjacency matrix.

[0019] A second aspect of the present invention provides a brain disease classification system based on heterogeneous perceptual spectral convolutional networks.

[0020] A brain disease classification system based on heterogeneous perceptual spectral convolutional networks, comprising: The crowd graph construction module is configured to: extract brain functional connectivity based on the resting-state brain functional images of the subjects, calculate the Pearson correlation between brain functional connectivity, and obtain the node features of the crowd graph; and construct the weights of the edges between nodes using a phenotypic encoder according to the relationships between different subjects, in order to construct the crowd graph. The classification module is configured as follows: based on the population graph, it calculates the local similarity between a node and its neighboring nodes to obtain the local similarity weights of all nodes; based on the local similarity weights of all nodes, it uses an MLP layer to obtain the weights of different layers, which are used to guide the feature fusion between layers to obtain the node feature representations of all layers; it uses a simplified spectral graph convolution to concatenate the feature representations of all layers to obtain multi-layer feature representations; based on the multi-layer feature representations, it obtains the brain disease classification probability matrix, thus obtaining the classification result.

[0021] A third aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in the first aspect above.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute steps in the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in the first aspect above.

[0023] The fifth aspect of the present invention provides a computer program product or computer program.

[0024] This invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in the first aspect above.

[0025] Compared with the prior art, the beneficial effects of the present invention are: This invention addresses the problem that nonlinear and heterogeneous relationships between individuals are difficult to model using static rules. It designs a phenotypic encoder to perform high-order interactive modeling of the phenotypic information of node pairs, dynamically generate edge weights, and achieve adaptive optimization of the population graph structure.

[0026] This invention designs a spectral graph convolutional network model with heterogeneity perception capabilities. Addressing the over-reliance of existing methods on homogeneity assumptions in heterogeneous scenarios, a local similarity learning mechanism is proposed. This mechanism measures the representational similarity between a node and its neighbors based on local topological information of the graph structure, and introduces this similarity as a homogeneity weight into the node aggregation process, achieving cross-level feature weighted fusion. Through this heterogeneity perception mechanism, the model can adapt to complex neighborhood structures with inconsistent categories in real-world population graphs, thereby significantly improving the classification accuracy and robustness of brain diseases. Attached Figure Description

[0027] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0028] Figure 1 This is a flowchart illustrating a brain disease classification method based on a heterogeneous perceptual spectral convolutional network, as shown in an embodiment of the present invention. Figure 2 This is a flowchart illustrating another embodiment of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks, as shown in this invention. Figure 3 This is a flowchart illustrating the dynamic construction of a crowd map based on a phenotypic encoder, as shown in an embodiment of the present invention. Figure 4 This is a structural diagram of a brain disease classification system based on a heterogeneous perceptual spectral convolutional network, as shown in an embodiment of the present invention. Figure 5 This is a structural diagram of a computer device shown in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0032] Figure 1 This is a flowchart illustrating a brain disease classification method based on heterogeneous perceptual spectral convolutional networks, as shown in an embodiment of the present invention; see reference. Figure 1 The method includes: Based on the resting-state brain function images of the subjects, brain functional connectivity was extracted, and Pearson correlations between brain functional connectivity were calculated to obtain the node features of the population graph. According to the relationships between different subjects, a phenotypic encoder was used to construct the weights of the edges between nodes to construct the population graph. Based on the population graph, the local similarity between a node and its neighboring nodes is calculated to obtain the local similarity weights of all nodes. Based on the local similarity weights of all nodes, an MLP layer is used to obtain the weights of different layers, which are used to guide the feature fusion between layers to obtain the node feature representations of all layers. Simplified spectral graph convolution is used to concatenate the feature representations of all layers to obtain multi-layer feature representations. Based on the multi-layer feature representations, the brain disease classification probability matrix is ​​obtained, and the classification results are obtained.

[0033] This invention provides a brain disease classification method based on a heterogeneous perceptual spectral graph convolutional network. It employs a population graph construction module to model the complex relationships between individuals, and then utilizes a heterogeneous perceptual spectral graph convolutional network to learn features from the constructed population graph, thereby fully capturing the nonlinear connections between individuals. Finally, based on the node classification results output by the graph network model, intelligent classification of brain diseases is achieved.

[0034] Figure 2 This is a flowchart illustrating another embodiment of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks, as shown in this invention; see also... Figure 2 This invention represents the crowd map as G = {V, }, where V is the set of nodes and A is the weight matrix of edges obtained through dynamic learning. First, the resting-state brain function images of the subjects are preprocessed to obtain brain functional connectivity. Then, the Pearson correlation between brain functional connectivity is calculated to obtain the node features in the graph.

[0035] To accurately model the potentially complex nonlinear relationships between node pairs, a phenotypic encoder is introduced to generate edge weights in the crowd graph. To dynamically model the complex relationships between individuals using nonlinear learning, this invention proposes a phenotypic encoder that dynamically assigns edge weights in the crowd graph by encoding pairs of phenotypic features. Specifically, individual subjects... and individuals The edge weights between them can be defined as: (1) in, and This is a multi-level mapping function, whose function is to map phenotypic features separately. and This mapping is performed to the hidden feature space. This mapping is achieved through a Multi-Layer Perceptron (MLP), a Rectified Linear Unit (ReLU), Batch Normalization (BN), Dropout, and another Multi-Layer Perceptron layer. The specific structure of the phenotypic encoder is as follows: Figure 3 As shown. To ensure consistency in feature processing, the weight parameters of the two mapping functions are shared. It is a bilinear weight matrix used to model the transformed feature representation in the hidden space. and The bilinear form captures higher-order interactions between paired features, significantly improving the model's ability to model complex relationships. Bias term. This further enhances the flexibility of the bilinear transform, and... The synergistic effect enables efficient modeling of feature correlations. This design allows subsequent graph neural networks to extract more representative and richer group-level features, laying the foundation for subsequent analysis.

[0036] Heterogeneity-aware spectral graph convolutional networks mainly include: local similarity learning mechanisms and simplified spectral graph convolution (S). 2The Local Similarity Learning (GC) module. The local similarity learning mechanism assigns appropriate weights to nodes based on their homogeneity, promoting more accurate and effective feature fusion between different simplified spectral graph convolutional modules. The simplified spectral graph convolutional module introduces a weighted summation mechanism that aggregates information from neighbors with different hop counts, preserving the original feature information of nodes while capturing the graph structure. The following sections will first elaborate on the local similarity learning mechanism and its theoretical analysis, followed by an introduction to the implementation details of the simplified spectral graph convolutional module.

[0037] (1) Learning mechanism based on local similarity Learning mechanisms based on local similarity measure the homogeneity of nodes by learning the similarity between a node and its neighboring nodes. Assuming that nodes with similar characteristics may belong to the same category, this invention first addresses the homogeneity of nodes... Define a simple local similarity To represent their homogeneity, the formula is as follows: (2) Where sim(·,·) is a similarity measurement function, such as cosine similarity or Euclidean similarity: Or (3) Formula (3) defines local similarity by calculating the average feature similarity between a node and its neighboring nodes. If cosine similarity or Euclidean similarity is used, the metric is linear and may not adequately represent the complex relationships between nodes. Therefore, this invention introduces a nonlinear term (i.e., This data is then processed using a multilayer perceptron to encode the distance variance between neighboring nodes. The formula for calculating the variance is as follows: (4) This variance reflects the degree of dispersion in the distribution of neighboring nodes. The precise definition of local similarity is as follows: (5) in, This represents a two-layer multilayer perceptron, with distance as the input feature. and its square By introducing a nonlinear term, It can capture true "local similarity" more accurately than before. More representative. Ultimately, this invention obtains the local similarity weights of all nodes, expressed as follows: (6) Local similarity weight This is then used to guide feature fusion between different layers in HA-SGCN, ultimately generating node representations and prediction results. Specifically, and its square Input into In the process, k weights are generated, each corresponding to a different layer. arrive These weights are used to guide feature fusion between layers: (7) Therefore, the node representation output by the k-th layer graph convolution is... It can be calculated as follows: (8) in, Indicates the k-th S 2 The output of the GC module. Finally, the node representations of all layers (including the initial representation) are compiled. The features are concatenated to obtain the final multi-layer feature representation: (9) Finally, this feature representation is passed through an MLP layer to obtain the final classification probability matrix. : (10) in, This represents the weights learned by the MLP layer.

[0038] (2) Simplified spectral convolution module Each simplified spectral convolution module comprises four components: a simplified spectral convolutional layer, a batch normalization layer, a ReLU activation layer, and a Dropout layer. This design not only enhances nonlinear expressiveness but also effectively mitigates overfitting. The core of simplified spectral convolution is to aggregate information from neighborhoods with different hop counts by introducing a weighted summation mechanism. This mechanism can capture the global graph structure while preserving the original feature information of each node, thus balancing the expressive power of both local and global information. Specifically, the... Layer (of which) The simplified definition of spectral convolution is as follows: (11) Here, L represents the number of times neighbor information is aggregated in the graph, i.e., the number of hops in the propagation process. It controls the scope of node information propagation in the graph structure. It is the learned weight matrix, used to perform linear transformations on the features. Indicates the process l The characteristics of jump propagation. It is the normalized propagation matrix of the feature, defined as follows: (12) Among them, through the adjacency matrix Adding self-loops can better capture the self-feature information of nodes. Is with The corresponding degree matrix is ​​defined as follows: (13) in, It is the matrix after adding self-loops to the adjacency matrix. In formula (11), It is a jump or restart probability ranging from [0, 1]. It is used to control the balance between utilizing the original features of nodes and relying on neighborhood structure information in the model. When the size is large, the model tends to preserve the original feature information of the nodes and reduce the modification of features during propagation; when When the size is small, the model focuses more on capturing graph structure through neighborhood information, thereby making better use of global information in the graph.

[0039] Unlike existing methods that use static linear similarity to define edge weights in crowd graphs, this invention proposes a dynamic graph construction strategy based on phenotypic encoder learning to build more adaptive and expressive crowd graphs, thereby capturing complex nonlinear relationships within the population. Specifically, this invention designs a phenotypic encoder that dynamically adjusts edge weights by encoding paired phenotypic features (such as age, gender, and collection location) to achieve more accurate graph structure construction. By dynamically modeling the complex nonlinear relationships between phenotypic features using the phenotypic encoder, the crowd graph can more accurately represent multidimensional information interactions between individuals. This method not only improves the model's adaptability to brain disease classification tasks but also provides a more accurate and efficient solution for the fusion and application of multimodal data.

[0040] Furthermore, addressing the limitations of relying on the homogeneity assumption in graph neural network modeling, this invention proposes a local similarity learning mechanism. This mechanism assesses the homogeneity level by measuring the similarity between a node and its neighboring nodes using local topological information. This mechanism integrates the homogeneity index into the node representation weighting process, achieving accurate fusion of cross-layer features in the graph network model. This further improves the accuracy of brain disease classification, increasing the accuracy of brain disease population classification by more than 5% compared to the latest methods.

[0041] The above combination Figure 1 The brain disease classification method based on heterogeneous perceptual spectral convolutional networks provided in the embodiments of the present invention has been described in detail. Next, the brain disease classification system based on heterogeneous perceptual spectral convolutional networks provided in the embodiments of the present invention will be described in conjunction with the accompanying drawings.

[0042] Figure 4 This is a schematic diagram of the structure of a brain disease classification system based on a heterogeneous perceptual spectral convolutional network, as shown in an embodiment of the present invention. (Refer to...) Figure 4 The system described in this invention includes: The crowd graph construction module is configured to: extract brain functional connectivity based on the resting-state brain functional images of the subjects, calculate the Pearson correlation between brain functional connectivity, and obtain the node features of the crowd graph; and construct the weights of the edges between nodes using a phenotypic encoder according to the relationships between different subjects, in order to construct the crowd graph. The classification module is configured as follows: Based on the crowd graph, it calculates the local similarity between a node and its neighboring nodes to obtain the local similarity weights of all nodes; based on the local similarity weights of all nodes, it uses an MLP layer to obtain the weights of different layers, which are used to guide feature fusion between layers to obtain the node feature representations of all layers; it uses simplified spectral graph convolution to concatenate the feature representations of all layers to obtain multi-layer feature representations; based on the multi-layer feature representations, it obtains the brain disease classification probability matrix, thus obtaining the classification result. In some embodiments, the weights of edges between nodes are constructed using a phenotypic encoder based on the relationships between different subjects; expressed by the following formula:

[0043] in, Indicates individual subjects and individuals Edge weights between them and This is a multi-level mapping function, whose function is to map phenotypic features separately. and Mapped to the hidden feature space; It is a bilinear weight matrix. This indicates the bias term.

[0044] In some embodiments, the step of calculating the local similarity between a node and its neighboring nodes based on the crowd graph to obtain the local similarity weights of all nodes includes: based on the crowd graph, introducing a nonlinear term and processing the nonlinear term through a multilayer perceptron, encoding the distance variance between neighboring nodes to obtain the local similarity between a node and its neighboring nodes, and combining the local similarities of all nodes to obtain the local similarity weights of all nodes.

[0045] In some embodiments, the step of using an MLP layer based on the local similarity weights of all nodes to obtain weights for different layers, which are used to guide feature fusion between layers to obtain node feature representations for all layers, includes: inputting the local similarity weights of all nodes and their squares into an MLP layer to generate weights for different layers, and then using the weights of each layer and each S... 2 The output of the GC module yields the node feature representations for each layer.

[0046] In some embodiments, the simplified spectral convolution is used to concatenate the feature representations of all layers, as expressed by the following formula:

[0047] in, Indicates the k-th S 2 The output of the GC module; L represents the number of times neighbor information is aggregated in the graph, i.e., the number of hops in the propagation process; It is the learned weight matrix; Indicates the process l The feature propagation matrix of the jump.

[0048] In some embodiments, the feature propagation matrix is ​​represented by the following formula:

[0049]

[0050] in, Represents the adjacency matrix. Is with The corresponding degree matrix, It is the matrix after adding self-loops to the adjacency matrix.

[0051] The brain disease classification system based on heterogeneous perceptual spectral convolutional networks according to embodiments of the present invention can correspond to performing the methods described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the brain disease classification system based on heterogeneous perceptual spectral convolutional networks are respectively for implementing Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0052] See Figure 5 The diagram shows the structure of a computer device, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium are connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory. The computer-readable storage medium stores computer programs, including program instructions, and the processor executes the program instructions stored in the computer-readable storage medium. The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device, adapted to implement one or more instructions, specifically adapted to load and execute one or more instructions to implement the corresponding steps in the embodiment of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks.

[0053] This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device. Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0054] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above embodiment of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks.

[0055] This embodiment provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps in the above embodiment of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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, generate instructions 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.

[0058] 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.

[0059] 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.

[0060] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A brain disease classification method based on heterogeneous perceptual spectral convolutional networks, characterized in that, include: Based on the resting-state brain function images of the subjects, brain functional connectivity was extracted, Pearson correlations between brain functional connectivity were calculated, and node features of the population map were obtained. Based on the relationships between different subjects, a phenotypic encoder is used to construct the weights of the edges between nodes in order to build a population graph; Based on the crowd graph, the local similarity between a node and its neighboring nodes is calculated to obtain the local similarity weights of all nodes; Based on the local similarity weights of all nodes, an MLP layer is used to obtain the weights of different layers, which are used to guide the feature fusion between layers to obtain the node feature representations of all layers. A simplified spectral convolution is used to concatenate the feature representations of all layers to obtain a multi-layer feature representation; based on the multi-layer feature representation, a brain disease classification probability matrix is ​​obtained, and the classification result is obtained.

2. The brain disease classification method based on heterogeneous perceptual spectral convolutional networks according to claim 1, characterized in that, The weights of edges between nodes are constructed using a phenotypic encoder based on the relationships between different subjects. The following formula is used to express this: in, Indicates individual subjects and individuals Edge weights between them and This is a multi-level mapping function, whose function is to map phenotypic features separately. and Mapped to the hidden feature space; It is a bilinear weight matrix. This indicates the bias term.

3. The brain disease classification method based on heterogeneous perceptual spectral convolutional networks according to claim 1, characterized in that, The method for calculating the local similarity between a node and its neighboring nodes based on a crowd graph, and obtaining the local similarity weights of all nodes, includes: introducing a nonlinear term based on the crowd graph and processing the nonlinear term through a multilayer perceptron to encode the distance variance between neighboring nodes, obtaining the local similarity between a node and its neighboring nodes, and combining the local similarities of all nodes to obtain the local similarity weights of all nodes.

4. The brain disease classification method based on heterogeneous perceptual spectral convolutional networks according to claim 1, characterized in that, The method involves using MLP layers to obtain weights for different layers based on the local similarity weights of all nodes, which guide feature fusion between layers to obtain node feature representations for all layers. The method includes: inputting the local similarity weights and their squares of all nodes into an MLP layer to generate weights for different layers; and then, based on the weights of each layer and each S... 2 The output of the GC module yields the node feature representations for each layer.

5. The brain disease classification method based on heterogeneous perceptual spectral convolutional networks according to claim 1, characterized in that, The simplified spectral graph convolution is used to concatenate the feature representations of all layers, as expressed by the following formula: in, Indicates the k-th S 2 The output of the GC module; L represents the number of times neighbor information is aggregated in the graph, i.e., the number of hops in the propagation process; It is the learned weight matrix; Indicates the process l The feature propagation matrix of the jump.

6. The brain disease classification method based on heterogeneous perceptual spectral convolutional networks according to claim 5, characterized in that, The feature propagation matrix is ​​expressed by the following formula: in, Represents the adjacency matrix. Is with The corresponding degree matrix, It is the matrix after adding self-loops to the adjacency matrix.

7. A brain disease classification system based on heterogeneous perceptual spectral convolutional networks, characterized in that, include: The crowd graph construction module is configured to: extract brain functional connectivity based on the resting-state brain functional images of the subjects, calculate the Pearson correlation between brain functional connectivity, and obtain the node features of the crowd graph; and construct the weights of the edges between nodes using a phenotypic encoder according to the relationships between different subjects, in order to construct the crowd graph. The classification module is configured to: calculate the local similarity between a node and its neighboring nodes based on the crowd graph, and obtain the local similarity weights of all nodes; based on the local similarity weights of all nodes, use an MLP layer to obtain the weights of different layers, which are used to guide the feature fusion between layers to obtain the node feature representations of all layers. A simplified spectral convolution is used to concatenate the feature representations of all layers to obtain a multi-layer feature representation; based on the multi-layer feature representation, a brain disease classification probability matrix is ​​obtained, and the classification result is obtained.

8. A computer device, characterized in that, A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the steps of the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the brain disease classification method based on heterogeneous perceptual spectral convolutional networks as described in any one of claims 1-6.

Images