Depression detection method, system, equipment and medium

Abstract

The invention discloses a depression detection method, system and device and a medium, and the method comprises the steps: obtaining electroencephalogram data to be subjected to depression detection; clustering graph construction is conducted on the electroencephalogram data, a clustering module graph is obtained, the clustering module graph comprises a plurality of clustering module nodes, and each clustering module node is used for representing a plurality of electroencephalogram signal channels with the same clustering center in the electroencephalogram data; carrying out topological graph construction on the electroencephalogram data to obtain a topological representation graph; according to the clustering module graph, performing multi-layer graph interactive fusion on the topological representation graph to obtain interactive fusion graph features; and carrying out feature classification on the interactive fusion image features to obtain a depression detection result of the electroencephalogram data. The method can improve the accuracy and adaptability of depression detection. The invention relates to the technical field of electroencephalogram data analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram data analysis, and in particular to a depression detection method, system, device and medium. Background Art

[0002] Depression is a common mental illness. Since early detection of depression is beneficial to timely intervene in patients with depression to control or reduce the impact of this mental illness on the physical and mental health of patients, depression detection has become one of the key concerns of people.

[0003] Currently, existing depression detection methods usually rely on predefined fixed brain region partitions, group electroencephalogram (EEG) channels into specific modules, and then achieve depression detection by analyzing the functional connections between and within the modules. However, due to the individual differences of patients with depression, there are dynamically changing brain network characteristics, and the detection accuracy of this method is not high, and the adaptability of depression detection is not satisfactory.

[0004] Therefore, the problems existing in the prior art still need to be solved and optimized urgently. Summary of the Invention

[0005] An object of the present invention is to solve at least to some extent one of the technical problems existing in the related art.

[0006] To this end, an object of an embodiment of the present invention is to provide a depression detection method, system, device and medium, wherein the method can improve the accuracy and adaptability of depression detection.

[0007] In order to achieve the above technical object, the technical solutions adopted in the embodiments of the present application include:

[0008] In a first aspect, an embodiment of the present application provides a depression detection method, including:

[0009] Obtain electroencephalogram data to be detected for depression;

[0010] Construct a clustering graph for the electroencephalogram data to obtain a clustering module graph, where the clustering module graph includes a plurality of clustering module nodes, and each clustering module node is used to represent a plurality of electroencephalogram signal channels with the same clustering center in the electroencephalogram data;

[0011] Construct a topological graph for the electroencephalogram data to obtain a topological representation graph, where the topological representation graph includes a plurality of topological representation nodes, and each topological representation node corresponds to an electroencephalogram signal channel in the electroencephalogram data;

[0012] According to the clustering module graph, perform multi-layer graph interaction fusion on the topological representation graph to obtain interaction fusion graph features;

[0013] Classify the features of the interactive fusion graph to obtain the depression detection result of the EEG data.

[0014] In addition, according to the method of the above embodiments of the present application, the following additional technical features may also be included:

[0015] Further, in an embodiment of the present application, the construction of the clustering graph for the EEG data to obtain the clustering module graph includes:

[0016] Perform basic spatial encoding on the EEG data to obtain basic view node embeddings;

[0017] Perform robust spatial encoding on the EEG data to obtain robust view node embeddings;

[0018] According to the basic view node embeddings, perform module fusion clustering on the robust view node embeddings to obtain the clustering module graph.

[0019] Further, in an embodiment of the present application, the performing module fusion clustering on the robust view node embeddings according to the basic view node embeddings to obtain the clustering module graph includes:

[0020] According to the basic view node embeddings, perform view feature fusion on the robust view node embeddings to obtain cross-view node embeddings;

[0021] Perform view module clustering on the cross-view node embeddings to obtain a number of clustering clusters, and each clustering cluster includes a number of cluster sample points, and each cluster sample point is used to represent an EEG signal channel in the EEG data;

[0022] Perform module graph conversion on all the clustering clusters to obtain the clustering module graph.

[0023] Further, in an embodiment of the present application, the performing multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain the interactive fusion graph features includes:

[0024] Perform multi-layer global feature extraction on the topological representation graph to obtain a first global feature and a number of second global features. The first global feature is the global topological feature of the topological representation graph at the last feature level, and the second global feature is any global topological feature other than the first global feature, and the feature levels of each second global feature are different;

[0025] According to all the second global features, perform multi-layer module feature extraction on the clustering module graph to obtain target module features;

[0026] According to the target module features, perform feature fusion on the first global feature to obtain the interactive fusion feature.

[0027] Further, in an embodiment of the present application, the obtaining of the target module features by performing multi-layer module feature extraction on the clustering module graph according to all the second global features includes:

[0028] Obtain the clustering matrix of the clustering module graph;

[0029] According to the clustering matrix, perform feature interaction on all the second global features to obtain intermediate interaction features corresponding to each of the second global features;

[0030] According to all the intermediate interaction features, perform multi-layer feature extraction on the clustering module graph to obtain the target module features.

[0031] Further, in an embodiment of the present application, the obtaining of the target module features by performing multi-layer feature extraction on the clustering module graph according to all the intermediate interaction features includes:

[0032] Obtain the first intermediate module feature of the clustering module graph at the current feature level;

[0033] Perform single-layer module feature extraction on the first intermediate module feature to obtain a second intermediate module feature;

[0034] According to the second intermediate module feature, perform feature selection on all the intermediate interaction features to obtain target interaction features, where the target interaction features are intermediate interaction features at the same feature level as the second intermediate module feature;

[0035] According to the target interaction features, perform feature splicing on the second intermediate module feature to obtain a third intermediate module feature;

[0036] If the feature level of the third intermediate module feature is not the last feature level of the clustering module graph, then update the first intermediate module feature according to the third intermediate module feature, and then return to execute the step of obtaining the first intermediate module feature of the clustering module graph at the current feature level; or, if the feature level of the third intermediate module feature is the last feature level of the clustering module graph, then determine the third intermediate module feature as the target module feature.

[0037] Further, in an embodiment of the present application, the obtaining of the depression detection result of the electroencephalogram data by performing feature classification on the interactive fusion graph feature includes:

[0038] Perform feature flattening on the interactive fusion graph feature to obtain a fused flattened feature;

[0039] Perform feature dimensionality reduction on the fused and flattened features to obtain fused and dimension-reduced features;

[0040] Perform feature mapping and classification on the fused and dimension-reduced features to obtain the depression detection result.

[0041] In a second aspect, an embodiment of the present application provides a depression detection system, including:

[0042] A first processing unit, configured to obtain electroencephalogram data to be detected for depression;

[0043] A second processing unit, configured to construct a clustering graph for the electroencephalogram data to obtain a clustering module graph, where the clustering module graph includes a plurality of clustering module nodes, and each clustering module node is used to represent a plurality of electroencephalogram signal channels in the electroencephalogram data that have the same clustering center;

[0044] A third processing unit, configured to construct a topological graph for the electroencephalogram data to obtain a topological representation graph, where the topological representation graph includes a plurality of topological representation nodes, and each topological representation node corresponds to an electroencephalogram signal channel in the electroencephalogram data;

[0045] A fourth processing unit, configured to perform multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain interaction fusion graph features;

[0046] A fifth processing unit, configured to perform feature classification on the interaction fusion graph features to obtain the depression detection result of the electroencephalogram data.

[0047] In a third aspect, an embodiment of the present application further provides an electronic device, including:

[0048] At least one processor;

[0049] At least one memory, configured to store at least one program;

[0050] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0051] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which a processor-executable program is stored, and the processor-executable program is used to implement the above method when executed by the processor.

[0052] The advantages and beneficial effects of the present application will be partially given in the following description, partially will become obvious from the following description, or be understood through the practice of the present application:

[0053] A depression detection method, system, device and medium disclosed in an embodiment of the present application. The method obtains electroencephalogram (EEG) data to be detected for depression; constructs a clustering graph for the EEG data to obtain a clustering module graph, where the clustering module graph includes a plurality of clustering module nodes, and each clustering module node is used to represent a plurality of EEG signal channels in the EEG data that have the same clustering center; constructs a topological graph for the EEG data to obtain a topological representation graph, where the topological representation graph includes a plurality of topological representation nodes, and each topological representation node corresponds to an EEG signal channel in the EEG data; performs multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain an interaction fusion graph feature; and classifies the interaction fusion graph feature to obtain a depression detection result of the EEG data. By constructing a clustering module graph of the EEG data, each clustering module node in the clustering module graph represents a plurality of EEG signal channels in the EEG data that have the same clustering center, which can dynamically construct a brain network module structure adapted to each different individual, is beneficial to alleviating the individual difference problem of depression patients, and further improves the detection accuracy and adaptability of depression detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following introduces the accompanying drawings related to the technical solutions in the embodiments of the present application or the prior art. It should be understood that the accompanying drawings below are only for conveniently and clearly expressing some embodiments of the technical solutions in the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a schematic flowchart of a depression detection method provided by an embodiment of the present application;

[0056] Figure 2 It is a schematic structural framework diagram of a depression detection system provided by an embodiment of the present application;

[0057] Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] Embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and should not be construed as a limitation of the present application. For the step numbers in the following embodiments, they are only set for the convenience of explanation and illustration, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0060] Currently, existing depression detection methods usually rely on predefined fixed brain region partitions, group electroencephalogram (EEG) channels into specific modules, and then achieve depression detection by analyzing the functional connections between and within the modules. However, due to the individual differences of depression patients, there are dynamic brain network characteristics, and it is difficult to accurately reflect the individualized brain function characteristics of different depression patients. The detection accuracy of this method is not high, and the adaptability of depression detection is not satisfactory. Moreover, this method also ignores the significant dynamic changes of depression patients in terms of time and task states, and there will be a certain loss of key information in the constructed brain network modules.

[0061] In addition, existing methods focus on analyzing this modular layer of brain network modules, lacking the fusion of global and modular information and multi-level interactions, unable to comprehensively capture complex brain network connection patterns, and thus making the accuracy and robustness of subsequent depression detection unsatisfactory.

[0062] In view of this, embodiments of the present invention provide a depression detection method, system, device, and medium. Among them, this method constructs a clustering module graph of EEG data. Each clustering module node in this clustering module graph represents a number of EEG signal channels with the same clustering center in the EEG data. This clustering module graph can not only adapt to the individual differences of depression patients, but also dynamically construct a brain network module structure suitable for each different individual, which is beneficial to alleviating the problem of individual differences of depression patients; at the same time, this clustering module graph can also capture the significant dynamics of brain network modules in terms of time and task states, which is beneficial to avoiding the loss of key information, and thus effectively improving the detection accuracy and adaptability of depression detection.

[0063] In addition, this method performs multi-layer graph interaction fusion on the clustering module graph and the topological representation graph. Specifically, it performs multi-layer global feature extraction on the topological representation graph, which can extract the topological relationships between various channels of EEG data from a global scope, thereby capturing the overall brain network characteristics of the EEG data. Moreover, based on the second global features at different feature levels, it performs multi-layer module feature extraction on the clustering module graph. While fully capturing the connection characteristics within and between modules, it also injects global features into the module features through an information interaction mechanism, thereby realizing the effective fusion of global features and module features, significantly improving the hierarchical expression ability of features, avoiding information redundancy and feature loss, and being beneficial to improving the accuracy and robustness of subsequent depression detection.

[0064] Referring to Figure 1 , in the embodiments of the present application, a depression detection method includes:

[0065] Step 110: Obtain EEG data to be detected for depression;

[0066] In the embodiments of the present application, the EEG data is the brain electrical activity data recorded by electroencephalography (EEG). This EEG data can be obtained by placing electrodes on the scalp to capture the electrical signals of brain cells, and it can be constructed from the EEG signals collected by a number of electrodes. Each electrode corresponds to an EEG signal channel. In addition, after obtaining the original EEG data, operations such as data augmentation can also be performed on the EEG data to obtain the EEG data after data augmentation. There are already various specific data augmentation methods. For example, based on a preset time length, the EEG data can be segmented into a number of non-overlapping EEG segments, and all the non-overlapping EEG segments after segmentation are determined as the EEG data after data augmentation.

[0067] Step 120: Construct a clustering graph for the EEG data to obtain a clustering module graph, where the clustering module graph includes a number of clustering module nodes, and each clustering module node is used to represent a number of EEG signal channels in the EEG data that have the same clustering center;

[0068] In the embodiments of the present application, a deep graph clustering network can be constructed, and the EEG data is input into the deep graph clustering network. The deep graph clustering network dynamically aggregates each EEG signal channel in the EEG data to form an adaptive brain network modular structure (i.e., the clustering module graph). The clustering module graph includes a number of clustering module nodes, and each clustering module node corresponds to a number of EEG signal channels in the EEG data that have the same clustering center.

[0069] In some embodiments, step 120: Construct a clustering graph for the EEG data to obtain a clustering module graph, includes:

[0070] A1. Perform basic spatial encoding on the electroencephalogram (EEG) data to obtain basic view node embeddings;

[0071] A2. Perform robust spatial encoding on the EEG data to obtain robust view node embeddings;

[0072] In the embodiments of the present application, the deep graph clustering network may include two non - shared encoders, namely a basic encoder and a robust encoder. The basic encoder is used to construct general node embeddings, and the robust encoder is used to introduce Gaussian perturbations to the node embeddings to construct robust node embeddings. Specifically, the basic encoder and the robust encoder may be multi - layer perceptron (MLP) networks, and the structures of the basic encoder and the robust encoder are the same but the parameters are different.

[0073] It can be understood that step A1 may be to input the EEG data into the basic encoder for basic spatial encoding, and the basic encoder captures the spatial embedding information of the EEG data to obtain basic view node embeddings; step A2 may be to input the EEG data into the robust encoder, and the robust encoder that introduces sampled Gaussian noise performs feature encoding on the EEG data to obtain robust view node embeddings.

[0074] A3. According to the basic view node embeddings, perform module fusion clustering on the robust view node embeddings to obtain the clustering module graph.

[0075] Further, step A3, according to the basic view node embeddings, perform module fusion clustering on the robust view node embeddings to obtain the clustering module graph, includes:

[0076] A31. According to the basic view node embeddings, perform view feature fusion on the robust view node embeddings to obtain cross - view node embeddings;

[0077] A32. Perform view module clustering on the cross - view node embeddings to obtain a number of clustering clusters, where each clustering cluster includes a number of cluster sample points, and each cluster sample point is used to represent an EEG signal channel in the EEG data;

[0078] A33. Perform module graph conversion on all the clustering clusters to obtain the clustering module graph.

[0079] In the embodiments of the present application, view feature fusion may be a high - dimensional fusion - based method to fuse the basic view node embeddings and the robust view node embeddings to obtain integrated cross - view node embeddings. Specifically, a linear combination operation may be used to perform high - dimensional fusion on the basic view node embeddings and the robust view node embeddings to obtain cross - view node embeddings.

[0080] It can be understood that step A32 can be to cluster the cross-view node embeddings using a clustering algorithm, mine the internal relationships of the implicit features of the nodes through the clustering algorithm, so as to obtain the clustering results of the cross-view node embeddings. The clustering results can include several clustering clusters, and each cluster sample point of the clustering cluster corresponds to an electroencephalogram (EEG) signal channel in the EEG data, which can enable the subsequent clustering module graph to generate a personalized brain network modular structure for the brain function characteristics of different individuals, times, and task states. Specifically, the clustering algorithm in the embodiments of the present application can specifically be any one of the K-Means clustering algorithm, the Mean-Shift clustering algorithm, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm, the hierarchical clustering algorithm, etc.

[0081] It should be noted that in practical applications, based on a clustering matrix, all clustering clusters in the clustering results can be represented by the indirect mapping relationships contained in the clustering matrix. For example, the clustering matrix includes several matrix elements, each matrix element is used to represent an EEG signal channel, and the element value on each matrix element is used to represent the assignment of the channel node of the EEG signal channel to the clustering cluster corresponding to the element value. Moreover, each clustering cluster can be used to represent a brain network module composed of several EEG signal channels, and the cluster sample point of the clustering cluster is the node within the brain network module; specifically, step A33 can be to convert each clustering cluster into the corresponding brain network module respectively, and construct a clustering module graph based on all brain network modules. In this clustering module graph, the node attributes or features within the brain network module are similar, and the node attributes or features between different brain network modules are significantly different.

[0082] It is worth mentioning that the embodiments of the present application can also introduce contrastive learning in the deep graph clustering network to update the assignment of EEG signal channels, so that the nodes in the same brain network module are closer in the embedding space, and the nodes in different brain network modules are more distant in the embedding space. Specifically, the view similarity matrix between two node embeddings can be calculated based on the basic view node embedding and the robust view node embedding, and then, based on the relationships contained in the clustering matrix, the corresponding relationship matrix is constructed. The deep graph clustering network is updated by restricting the view similarity matrix to approximate the relationship matrix, thereby further improving the applicability of the clustering module graph to individual differences, time, and dynamic changes in task states, which is beneficial to improving the accuracy of depression detection.

[0083] Step 130: Construct a topological graph for the EEG data to obtain a topological representation graph, where the topological representation graph includes several topological representation nodes, and each topological representation node corresponds to an EEG signal channel in the EEG data;

[0084] In the embodiments of the present application, each electroencephalogram (EEG) signal channel in the EEG data can be regarded as a node in the corresponding graph structure, and the relationship between the EEG signal channels can be represented by an adjacency matrix. Then, based on the adjacency matrix and each EEG signal channel, a topological representation graph corresponding to the EEG data is constructed, and the topological representation graph is used to provide a global graph topological representation.

[0085] Step 140: Perform multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain interaction fusion graph features;

[0086] In the embodiments of the present application, a graph interaction learning module can be constructed, and the clustering module graph and the topological representation graph are input into the graph interaction learning module. Through the graph interaction learning module, global graph learning is performed on the topological representation graph, and modular graph learning is performed on the clustering module graph based on an interaction mechanism, so as to obtain interaction fusion graph features.

[0087] In some embodiments, step 140: Perform multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain interaction fusion graph features, includes:

[0088] B1: Perform multi-layer global feature extraction on the topological representation graph to obtain a first global feature and several second global features. The first global feature is the global topological feature of the topological representation graph at the last feature level, and the second global feature is any global topological feature other than the first global feature. The feature levels of each second global feature are different, and the feature receptive field of the global topological feature at the subsequent feature level is larger than the feature receptive field of the global topological feature at the previous feature level;

[0089] In the embodiments of the present application, the global graph learning of the graph interaction learning module can be implemented by a global layer in the graph interaction learning module. The global layer includes several cascaded graph convolutional networks (GCNs). By the cascaded graph convolutional networks, the node features and neighbor node features of the topological representation graph are combined to capture the brain network characteristics between each EEG signal channel of the EEG data layer by layer, so as to obtain several global topological features at different feature levels. Then, the global topological feature at the last feature level is determined as the first global feature, and the global topological features at the remaining feature levels are respectively used as a second global feature.

[0090] Specifically, in the embodiment of the present application, taking the number of cascaded graph convolutional networks for realizing global graph learning as 3 as an example, the cascaded graph convolutional networks can be the first graph convolutional network, the second graph convolutional network, and the third graph convolutional network respectively. Among them, the input of the first graph convolutional network is the topological representation graph, and the output is the global topological feature of the first feature level (i.e., the second global feature of the first feature level). The input of the second graph convolutional network is the global topological feature of the first feature level, and the output of the second graph convolutional network is the global topological feature of the second feature level. The output of the third graph convolutional network is the first global feature.

[0091] B2. Perform multi-layer module feature extraction on the clustering module graph according to all the second global features to obtain target module features;

[0092] Further, the step B2. Perform multi-layer module feature extraction on the clustering module graph according to all the second global features to obtain target module features includes:

[0093] B21. Obtain the clustering matrix of the clustering module graph;

[0094] B22. Perform feature interaction on all the second global features according to the clustering matrix to obtain intermediate interaction features corresponding to each second global feature;

[0095] In the embodiment of the present application, the modular graph learning of graph interaction learning can be implemented by the module layer in the graph interaction learning module. The module layer includes several cascaded graph convolutional networks (GCNs). Through the module layer, each second global feature is extracted layer by layer with the intermediate module features of the corresponding feature level. It can capture the module features of the layer brain network while retaining the low-level detail information, so as to obtain the target module features of the clustering module graph and several intermediate interaction features. Among them, the target module features are the module features of the last feature level, and the intermediate interaction features are the module features except the last feature level. The feature receptive field of the module features of the latter feature level is larger than that of the module features of the previous feature level.

[0096] It can be understood that the clustering matrix of the clustering module graph can be the matrix representation corresponding to all the clustering clusters in the foregoing step A32; step B22 can be to perform in-module fusion on each second global feature based on the clustering matrix, so as to obtain intermediate interaction features corresponding to each second global feature, and each intermediate interaction feature corresponds to the module feature output by a graph convolutional network.

[0097] B23. Perform multi-layer feature extraction on the clustering module graph according to all the intermediate interaction features to obtain the target module features.

[0098] Further, step B23, extracting multi-layer features from the clustering module graph according to all the intermediate interaction features to obtain the target module features, includes:

[0099] B231. Obtain the first intermediate module feature of the clustering module graph at the current feature level;

[0100] B232. Perform single-layer module feature extraction on the first intermediate module feature to obtain a second intermediate module feature;

[0101] B233. According to the second intermediate module feature, perform feature selection on all the intermediate interaction features to obtain target interaction features, where the target interaction features are intermediate interaction features at the same feature level as the second intermediate module feature;

[0102] B234. According to the target interaction features, perform feature splicing on the second intermediate module feature to obtain a third intermediate module feature;

[0103] B235. If the feature level of the third intermediate module feature is not the last feature level of the clustering module graph, update the first intermediate module feature according to the third intermediate module feature, and then return to execute the step of obtaining the first intermediate module feature of the clustering module graph at the current feature level;

[0104] Alternatively, B236. If the feature level of the third intermediate module feature is the last feature level of the clustering module graph, determine the third intermediate module feature as the target module feature.

[0105] In the embodiments of the present application, the first intermediate module feature may be the input of a certain graph convolutional network in the module layer, and this input may be the clustering module graph or the third intermediate module feature output by the previous graph convolutional network. Step B232 may be to perform graph convolutional learning on the input first intermediate module feature through a graph convolutional network to obtain the second intermediate module feature output by this graph convolutional network.

[0106] It can be understood that after obtaining the second intermediate module feature, it is possible to perform feature concatenation on the second intermediate module feature and the intermediate interaction feature at the same feature level to obtain a third intermediate module feature, which is the final output of the module layer at the current feature level. Additionally, if the feature level of the current third intermediate module feature is not the last feature level of the clustering module graph, the current third intermediate module feature can be used as the first intermediate module feature at the next feature level, which can be specifically achieved by using the current third intermediate module feature as the input to the next graph convolutional network; alternatively, if the current third intermediate module feature is at the feature level of the clustering module, the current third intermediate module feature can be determined as the module feature at the last feature level (i.e., the target module feature).

[0107] Exemplarily, in an embodiment of the present application, it is assumed that the module layer includes three cascaded graph convolutional networks, which are the fourth graph convolutional network, the fifth graph convolutional network, and the sixth graph convolutional network in sequence. Among them, the feature level of the third intermediate module feature obtained after the sixth graph convolutional network is the last layer feature level. Specifically, first, the clustering module graph is used as the initial first intermediate module feature and input into the fourth graph convolutional network to obtain the second intermediate module feature at the first feature level output by the fourth graph convolutional network; then, the second intermediate module feature at the first feature level and the intermediate interaction feature at the corresponding feature level are concatenated to obtain the third intermediate module feature at the first feature level; next, the third intermediate module feature at the first feature level is used as the first intermediate module feature input into the fifth graph convolutional network, and after feature extraction, selection, and concatenation, the third intermediate module feature at the second feature level is obtained; the third intermediate module feature at the second feature level is used as the first intermediate module feature input into the sixth graph convolutional network, and after feature extraction, selection, and concatenation, the third intermediate module feature at the third feature level is obtained, and the third intermediate module feature at the third feature level is the target module feature.

[0108] B3. According to the target module feature, perform feature fusion on the first global feature to obtain the interaction fusion feature.

[0109] In an embodiment of the present application, after obtaining the first global feature output by the global layer and the target module feature output by the module layer, a feature fusion operation can be performed on the output first global feature and target module feature. There are already various implementation methods for the specific feature fusion operation. For example, it is possible to implement the feature fusion of the first global feature and the target module feature based on the attention mechanism, based on element-wise multiplication or element-wise addition operations, etc., to obtain the interaction fusion feature.

[0110] Step 150: Classify the interactive fusion graph features to obtain the depression detection result of the EEG data.

[0111] In the embodiment of the present application, after obtaining the interactive fusion graph features, the interactive fusion graph features can be input into a classifier for depression detection, so as to obtain the depression detection result of the EEG data.

[0112] In some embodiments, Step 150: Classify the interactive fusion graph features to obtain the depression detection result of the EEG data, includes:

[0113] C1. Flatten the interactive fusion graph features to obtain flattened fusion features;

[0114] C2. Reduce the dimension of the flattened fusion features to obtain reduced-dimension fusion features;

[0115] C3. Map and classify the reduced-dimension fusion features to obtain the depression detection result.

[0116] In the embodiment of the present application, the interactive fusion graph features can be input into a classifier. The flattening layer in the classifier flattens the interactive fusion graph features into a one-dimensional feature vector, and this one-dimensional feature vector is determined as the flattened fusion features; then, the two linear layers in the classifier are used to reduce the dimension of the flattened fusion features, so as to obtain the reduced-dimension fusion features; then, the Softmax function is used to map the reduced-dimension fusion features into classification labels, so as to obtain the depression detection result of the EEG data.

[0117] Next, a depression detection system proposed according to an embodiment of the present application will be described in detail with reference to the accompanying drawings.

[0118] Refer to Figure 2 , a depression detection system proposed in the embodiment of the present application, includes:

[0119] The first processing unit 101 is configured to obtain EEG data to be detected for depression;

[0120] The second processing unit 102 is configured to construct a clustering graph for the EEG data to obtain a clustering module graph, and the clustering module graph includes a plurality of clustering module nodes, and each clustering module node is used to represent a plurality of EEG signal channels in the EEG data that have the same clustering center;

[0121] The third processing unit 103 is configured to construct a topological graph for the EEG data to obtain a topological representation graph, and the topological representation graph includes a plurality of topological representation nodes, and each topological representation node corresponds to an EEG signal channel in the EEG data;

[0122] A fourth processing unit 104, configured to perform multi-layer graph interaction fusion on the topological representation graph according to the clustering module graph to obtain interaction fusion graph features;

[0123] A fifth processing unit 105, configured to perform feature classification on the interaction fusion graph features to obtain a depression detection result of the electroencephalogram data.

[0124] It can be understood that the content in the above method embodiments is applicable to the system embodiments. The functions specifically implemented in the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0125] Referring to Figure 3 , an embodiment of the present application further provides an electronic device, including:

[0126] At least one processor 201;

[0127] At least one memory 202, configured to store at least one program;

[0128] When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the above method embodiments.

[0129] Similarly, it can be understood that the content in the above method embodiments is applicable to the device embodiments. The functions specifically implemented in the device embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0130] An embodiment of the present application further provides a computer-readable storage medium, in which a program executable by a processor 201 is stored, and the program executable by the processor 201 is used to implement the above method embodiments when executed by the processor 201.

[0131] Similarly, the content in the above method embodiments is applicable to the computer-readable storage medium embodiments. The functions specifically implemented in the computer-readable storage medium embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0132] In some alternative embodiments, the functions / operations recited in the block diagrams may not occur in the order presented in the operational illustrations. For example, depending on the functions / operations involved, two blocks shown in succession may actually be executed substantially simultaneously or the blocks can sometimes be executed in the reverse order. Additionally, the embodiments presented and described in the flowcharts of the present application are provided by way of example for the purpose of providing a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and in which sub-operations described as part of a larger operation are executed independently.

[0133] Furthermore, although the present application has been described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated in a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It should also be understood that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present application. Rather, given the attributes, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the modules will be understood within the ordinary skill of an engineer. Thus, those of ordinary skill in the art will be able to implement the present application as set forth in the claims without undue experimentation. It should also be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the present application, which is determined by the full scope of the appended claims and their equivalents.

[0134] If a function is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or a part of this technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method according to the embodiments of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a portable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc.

[0135] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definitional sequence list of executable instructions for implementing logical functions, which can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0136] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which a program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing as necessary, and then stored in a computer memory.

[0137] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), and the like.

[0138] In the above description of this specification, the descriptions referring to the terms "one embodiment / example", "another embodiment / example", or "certain embodiments / examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0139] Although the embodiments of the present application have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present application. The scope of the present application is defined by the claims and their equivalents.

[0140] The above is a specific description of the preferred embodiments of the present application, but the present application is not limited to the embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present application, and these equivalent deformations or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A depression detection method, characterized in that: include: Obtaining EEG data to be tested for depression; Constructing a clustering graph for the EEG data to obtain a clustering module graph, wherein the clustering module graph includes a plurality of clustering module nodes, each of which is used to represent a plurality of EEG signal channels having the same clustering center in the EEG data; Constructing a topological graph of the EEG data to obtain a topological representation graph, wherein the topological representation graph includes a plurality of topological representation nodes, each of which corresponds to an EEG signal channel in the EEG data; According to the clustering module graph, the topology representation graph is interactively fused with multiple layers to obtain interactive fusion graph features; The interactive fusion graph features are classified to obtain depression detection results of the EEG data.

2. The method according to claim 1, characterized in that The clustering diagram of the EEG data is constructed to obtain a clustering module diagram, including: Performing basic spatial encoding on the EEG data to obtain basic view node embedding; Performing robust spatial encoding on the EEG data to obtain a robust view node embedding; According to the basic view node embedding, module fusion clustering is performed on the robust view node embedding to obtain the cluster module graph.

3. The method according to claim 2, characterized in that The performing module fusion clustering on the robust view node embedding according to the basic view node embedding to obtain the clustering module graph includes: According to the basic view node embedding, performing view feature fusion on the robust view node embedding to obtain a cross-view node embedding; Performing view module clustering on the cross-view node embedding to obtain a plurality of clusters, wherein the clusters include a plurality of cluster sample points, and each of the cluster sample points is used to represent an EEG signal channel in the EEG data; All the clusters are converted into module graphs to obtain the cluster module graphs.

4. The method according to claim 1, characterized in that The step of performing multi-layer interactive fusion of the topology representation graph according to the clustering module graph to obtain interactive fusion graph features includes: Performing multi-layer global feature extraction on the topological representation graph to obtain a first global feature and a plurality of second global features, wherein the first global feature is a global topological feature of the topological representation graph at the last feature level, and the second global feature is any global topological feature other than the first global feature, and each of the second global features has a different feature level; According to all the second global features, multi-layer module feature extraction is performed on the clustering module graph to obtain target module features; According to the target module feature, the first global feature is fused to obtain the interactive fusion feature.

5. The method according to claim 4, characterized in that The step of performing multi-layer module feature extraction on the cluster module graph according to all the second global features to obtain target module features includes: Obtaining a clustering matrix of the clustering module graph; According to the clustering matrix, performing feature interaction on all the second global features to obtain an intermediate interaction feature corresponding to each of the second global features; According to all the intermediate interaction features, multi-layer feature extraction is performed on the clustering module graph to obtain the target module features.

6. The method according to claim 5, characterized in that The step of performing multi-layer feature extraction on the clustering module graph according to all the intermediate interaction features to obtain the target module features includes: Obtaining a first intermediate module feature of the clustering module graph at a current feature level; Performing single-layer module feature extraction on the first intermediate module feature to obtain a second intermediate module feature; According to the second intermediate module feature, feature selection is performed on all the intermediate interaction features to obtain a target interaction feature, where the target interaction feature is an intermediate interaction feature at the same feature level as the second intermediate module feature; According to the target interaction feature, feature concatenation is performed on the second intermediate module feature to obtain a third intermediate module feature; If the feature level of the third intermediate module feature is not the last feature level of the clustering module graph, the first intermediate module feature is updated according to the third intermediate module feature, and then the step of obtaining the first intermediate module feature of the clustering module graph at the current feature level is returned; or, if the feature level of the third intermediate module feature is the last feature level of the clustering module graph, the third intermediate module feature is determined as the target module feature.

7. The method according to claim 1, characterized in that The performing feature classification on the interactive fusion graph features to obtain the depression detection result of the EEG data includes: Flattening the interactive fusion graph features to obtain fused flattened features; Performing feature dimensionality reduction on the fused flattened features to obtain fused dimensionality reduction features; Feature mapping and classification are performed on the fused dimensionality reduction features to obtain the depression detection result.

8. A depression detection system, characterized in that: include: A first processing unit, for acquiring EEG data to be detected for depression; A second processing unit is used to construct a clustering graph for the EEG data to obtain a clustering module graph, wherein the clustering module graph includes a plurality of clustering module nodes, and each of the clustering module nodes is used to represent a plurality of EEG signal channels having the same clustering center in the EEG data; A third processing unit is used to construct a topological graph for the EEG data to obtain a topological representation graph, wherein the topological representation graph includes a plurality of topological representation nodes, each of which corresponds to an EEG signal channel in the EEG data; A fourth processing unit is used to perform multi-layer interactive fusion on the topology representation graph according to the clustering module graph to obtain interactive fusion graph features; The fifth processing unit is used to perform feature classification on the interactive fusion image features to obtain a depression detection result of the EEG data.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images