Asynchronous brain network-based affective disorder identification method and system
The method addresses the challenge of emotion disorder identification by employing asynchronous brain network analysis and graph neural networks to enhance the accuracy and reliability of emotion disorder detection.
Patent Information
- Application Number
- CN202510499456.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-07-15
AI Technical Summary
The existing technology lacks accurate, reliable and highly generalized computer-aided recognition models, and fails to fully examine the complexity and dynamic evolution of the brain asynchronous information processing process, resulting in insufficient accuracy in recognition of emotional disorders.
Inhomogeneous EEG sampling is used to obtain asynchronous EEG signals, pre-processing such as filtering, artifact removal, frequency band decomposition and baseline drift correction, asynchronous topology features and channel time domain features are extracted, feature fusion is performed through graph neural network and Transformer model, emotional disorder recognition model is constructed, and optimization strategy training is adopted to improve accuracy.
A computer-assisted emotional disorder recognition model with highly accurate, reliable and powerful generalization capabilities has been developed, which significantly improves the accuracy of early judgment of emotional disorders.
Smart Images
Figure CN120304847A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of the integration of neuroscience and computer technology, and particularly to a method and system for identifying emotional disorders based on an asynchronous brain network. Background Art
[0002] With the development of neuroscience and brain imaging technology, it has become a key breakthrough point to explore objective indicators that can effectively characterize emotional disorders from the level of brain physiological and pathological mechanisms. As the core of cognitive and emotional regulation, the information processing process of the brain does not follow a synchronous and linear pattern, but there is a complex asynchronous processing mechanism.
[0003] Specifically, in the emotional process triggered by events, each brain region will perform asynchronous coordinated processing and integration of emotion-related information on different time scales. However, the current research on the internal relationship between the brain's asynchronous information processing mechanism and emotional disorders is still insufficient in depth and faces many theoretical and methodological challenges.
[0004] In the field of identifying emotional disorders based on this mechanism, there is still a lack of a computer-aided recognition model that is accurate, reliable, and has strong generalization ability. Although some existing studies use neuroimaging techniques such as electroencephalogram (EEG) signals and functional magnetic resonance imaging (fMRI) to construct synchronous brain network models, they only perform shallow analysis on the activities of the whole or part of the brain regions at the same moment of the brain, and fail to fully examine the highly complex and dynamically evolving delay characteristics in the brain's asynchronous information processing process. Summary of the Invention
[0005] The purpose of the present invention is to provide a method and system for identifying emotional disorders based on an asynchronous brain network, and through innovative theories and methods, develop a highly accurate, reliable, and powerful computer-aided emotional disorder recognition model to significantly improve the accuracy of early judgment of emotional disorders.
[0006] To achieve the above purpose, the present invention provides a method for identifying emotional disorders based on an asynchronous brain network, including the following steps:
[0007] S1: Perform non-uniform EEG sampling, obtain asynchronous EEG signals and resample them;
[0008] S2: Perform primary preprocessing and secondary preprocessing on the asynchronous EEG signals to obtain aligned and standardized signal segments Wherein, represents the data of the k-th time segment from the i-th EEG signal channel, I is the total number of channels, and K is the total number of time segments;
[0009] S3: Extract features from the aligned and standardized signal segments to obtain asynchronous topological features and channel time-domain features of the signal segments;
[0010] S4: Convert the asynchronous topological features and channel time-domain features of the signal segments into a unified format, and fuse them to obtain unified spatio-temporal fusion features;
[0011] S5: Construct a preliminary emotional disorder recognition model, set emotional disorder labels, input the spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and use an optimization strategy to train the preliminary emotional disorder recognition model to obtain the optimal emotional disorder model;
[0012] S6: Use the optimal emotional disorder model to predict the emotional disorder state and perform visual analysis.
[0013] Preferably, preprocess the asynchronous EEG signals to obtain aligned and standardized signal segments. The preprocessing includes filtering, rejection, decomposition, as well as alignment and standardization processing, where the alignment and standardization processing includes
[0014] Perform baseline drift correction on the asynchronous brain signals of each channel, use event trigger markers to perform time alignment on the asynchronous EEG signals of different channels, divide according to the event interval, perform time-domain alignment on the EEG signals within the event interval, and intercept equal-length sequences. Perform amplitude normalization or Z-score standardization on the time-domain aligned EEG signals to obtain aligned and standardized signal segments.
[0015] Preferably, the asynchronous topological features include: node degree, clustering coefficient, path length, betweenness centrality, eigenvector centrality, local efficiency, and global efficiency.
[0016] Preferably, set a first feature extraction module to extract the asynchronous topological features of the EEG signals, including:
[0017] Set a first extraction unit to extract features from the information segments to obtain local features;
[0018] Set a global information acquisition module for acquiring the global structural information of the information segments;
[0019] Set a first correlation analysis unit for predicting between local features according to the global structural information of the information segments, performing correlation analysis on the local features according to the prediction results, setting a first correlation threshold, and rejecting the local features below the first correlation threshold to obtain the asynchronous topological features.
[0020] Preferably, set a second feature extraction module to extract the channel time-domain features of the EEG signals, including:
[0021] Set a second extraction unit to extract time-domain features from the information segments according to the information segments to obtain first time-domain features;
[0022] Perform positional encoding on the first time-domain features, and set up a second correlation analysis unit to analyze the correlation of the first time-domain features with positional encoding, including the correlation of information segments at adjacent positions and the correlation of each information segment in the global information;
[0023] Set a second correlation threshold, and eliminate the first time-domain features with positional encoding that are lower than the second correlation threshold to obtain the second time-domain features, that is, channel time-domain features.
[0024] Preferably, the optimization strategy includes
[0025]
[0026] In the formula,
[0027] Among them, is the total loss function, which consists of two parts: the basic loss and the graph regularization loss λ is a hyperparameter that controls the weight of the graph regularization loss; in y i is the true class label of the i-th sample; is the model prediction probability of the i-th sample, in w ij is the weight matrix of the graph structure, representing the connection weight between nodes i and j; is the embedding of node i calculated by the graph neural network; is the embedding of node j calculated by the graph neural network; ||·|| 2 represents the square of the Euclidean norm, and calculates the squared distance between two node embedding vectors.
[0028] The emotional disorder recognition system based on the asynchronous brain network includes
[0029] A data acquisition module for performing non-uniform EEG sampling, obtaining asynchronous EEG signals and resampling them;
[0030] A data processing module for preprocessing the asynchronous EEG signals to obtain aligned and standardized signal segments Among them, represents the data of the k-th time segment from the i-th EEG signal channel, I is the total number of channels, and K is the total number of time segments;
[0031] A first feature extraction module for extracting features from the aligned and standardized signal segments to obtain the asynchronous topological features of the signal segments;
[0032] A second feature extraction module for extracting features from the aligned and standardized signal segments to obtain the channel time-domain features of the signal segments;
[0033] A feature fusion module, configured to convert the asynchronous topological features and channel time-domain features of signal segments into a unified format, and fuse them to obtain unified spatio-temporal fusion features;
[0034] A model construction module, configured to construct a preliminary emotional disorder recognition model, set emotional disorder labels, input the spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and train the preliminary emotional disorder recognition model using an optimization strategy to obtain an optimal emotional disorder model;
[0035] An analysis module, configured to predict the emotional disorder state using the optimal emotional disorder model and perform visual analysis.
[0036] Preferably, the first feature extraction module includes a first extraction unit and a first correlation analysis unit; the second feature extraction module includes a second extraction unit and a second correlation analysis unit.
[0037] Therefore, the present invention adopts the above-mentioned method and system for recognizing emotional disorders based on an asynchronous brain network, and through innovative theories and methods, develops a highly accurate, reliable and powerful computer-aided emotional disorder recognition model to significantly improve the accuracy of early judgment of emotional disorders. Brief Description of the Drawings
[0038] Figure 1 It is a schematic diagram of the emotional disorder recognition system based on an asynchronous brain network of the present invention. Detailed Embodiments
[0039] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.
[0040] Unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those of ordinary skill in the art to which the present invention belongs.
[0041] Embodiment 1
[0042] As Figure 1 shown, the method for recognizing emotional disorders based on an asynchronous brain network includes the following steps:
[0043] S1: Perform non-uniform EEG sampling, obtain asynchronous EEG signals and perform resampling;
[0044] Under the stimulation of different electroencephalogram (EEG) experimental procedures, using non-uniform sampling technology based on an event-driven A / D converter (EDADC), an experimental paradigm based on psychology and neuroscience is used to induce asynchronous information transmission in the brain, and corresponding high-density EEG data is acquired in an asynchronous event-triggered manner; the non-uniformly sampled EEG signals are resampled to be applicable to traditional signal processing tools and algorithms.
[0045] S2: Perform primary preprocessing and secondary preprocessing on the asynchronous EEG signals to obtain aligned and standardized signal segments Among them, represents the data of the k-th time segment from the i-th EEG signal channel, I is the total number of channels, and K is the total number of time segments;
[0046] Perform primary preprocessing on the resampled EEG signal data. The specific steps include:
[0047] Signal filtering: Filter the multi-channel EEG signal data to remove power frequency noise, electromyogram artifacts, and other interference signals, and obtain preliminarily denoised band-limited EEG signal data.
[0048] Artifact removal: Use the ICA algorithm for artifact detection, identify and remove physiological artifacts such as eye movements and electrocardiograms, so as to obtain pure EEG signal data.
[0049] Band decomposition: Decompose the denoised EEG signals into different frequency bands, extract rhythm signals in specific frequency ranges, such as δ (0.5 - 4 Hz), θ (4 - 8 Hz), α (8 - 13 Hz), β (13 - 30 Hz), and γ (above 30 Hz), etc., for subsequent analysis.
[0050] Divide the asynchronous EEG data after primary preprocessing according to the event interval, perform time domain alignment on the EEG signals within the event interval, and intercept sequences of the same length;
[0051] Preferably, perform secondary preprocessing on the EEG signal data of different channels after primary preprocessing by alignment and standardization, specifically including the following steps:
[0052] Baseline correction: Correct the baseline drift of each channel signal to ensure that the signals are analyzed at the same starting point.
[0053] Time synchronization: Use event trigger markers to perform time alignment on the EEG signals of different channels.
[0054] Signal segmentation: According to the experimental task or analysis requirements, perform fixed-duration time window segmentation on the aligned EEG signal data, for example, in a sliding window or fixed window manner, to ensure the temporal consistency of subsequent feature extraction.
[0055] Data standardization: Perform amplitude normalization or Z-score standardization on the signals after time-domain alignment to reduce the impact of individual differences on the analysis results and improve the robustness of the model.
[0056] S3: Extract asynchronous topological features and channel time-domain features of the signal segments from the aligned and standardized signal segments; the matrix can extract asynchronous topological features through the brain network model and the graph neural network model, and extract channel time-domain features through the transformer model.
[0057] The asynchronous EEG signals of each subject are represented by , and the corresponding asynchronous brain network is defined as a weighted graph G=(V,A,W,X), where:
[0058]
[0059]
[0060] The above sets respectively represent the node set I defined by the asynchronous brain network model; the adjacency matrix A, where A(i,j) represents the connection relationship between nodes i and j, the weight matrix W, where W(i,j) represents the weight of the edge between nodes i and j; the node feature matrix X, where X(i,d) represents the dth feature of the ith node. Each node vector x(i) contains the following features: node degree k i , clustering coefficient C i , average path length L of the node i , betweenness centrality, BC i eigenvector centrality EC i , local efficiency LE i , global efficiency GE i And serve as the input to the following graph neural network.
[0061] 2. The graph neural network extracts topological features
[0062] Graph feature update rule: After L layers of GNN, obtain the topological feature matrix.
[0063] 3. Transformer extracts temporal features:
[0064] This module is responsible for extracting time series features from asynchronous time series data, and finally obtaining the global temporal feature pooling output of the Transformer:
[0065] Set the first feature extraction module to extract asynchronous topological features of the EEG signals, including:
[0066] Set the first extraction unit to extract features from the information segments to obtain local features;
[0067] Set the global information acquisition module to obtain the global structural information of the information segments;
[0068] Set the first correlation analysis unit to predict between local features according to the global structural information of the information segments, perform correlation analysis on the local features according to the prediction structure, set the first correlation threshold, and eliminate the local features below the first correlation threshold to obtain asynchronous topological features.
[0069] Set the first correlation analysis unit to predict between local features according to the global structural information of the information segments, perform correlation analysis on the local features according to the prediction structure, including
[0070] Calculate the edge weight set W: Calculate the Pearson correlation coefficient between the segmented signals of each channel in the matrix to measure the linear correlation between channel i and channel j.
[0071] Construct the adjacency matrix set A: Use the brain regions corresponding to the EEG signal channels as network nodes, defined as V = {v1, v2,..., v I}, and use the correlation values in the Pearson correlation matrix as the weights of the edges to define the brain network adjacency matrix A(i, j).
[0072] Calculate the node features X: Calculate the node topological features and fuse them to obtain the node feature vector. The node topological features specifically include: node degree, clustering coefficient, node average path length, betweenness centrality, eigenvector centrality, local efficiency, global efficiency
[0073] Concatenate the calculated node features into the feature vector of node i For the node set V, the corresponding expression of the feature matrix is given: X = [v (1) , v (2) ,... v (I) .
[0074] Set the second feature extraction module to extract the channel time-domain features of the EEG signals, including:
[0075] Set the second extraction unit to extract the time-domain features from the information segments according to the information segments to obtain the first time-domain features;
[0076] Perform position encoding on the first time-domain features, and set the second correlation analysis unit to analyze the correlation of the first time-domain features with position encoding, including the correlation of adjacent information segments and the correlation of each information segment in the global information;
[0077] Set a second correlation threshold, and remove the first time-domain features containing position encoding that are lower than the second correlation threshold to obtain the second time-domain features, i.e., channel time-domain features.
[0078] S4: Convert the asynchronous topological features and channel time-domain features of the signal segment into a unified format and fuse them to obtain unified spatio-temporal fusion features; use an MLP layer to fuse the spatio-temporal features.
[0079] S5: Build a preliminary emotional disorder recognition model, set emotional disorder labels, input the spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and use an optimization strategy to train the preliminary emotional disorder recognition model to obtain the optimal emotional disorder model; the classifier uses the softmax function for output.
[0080] During the training process, cross-entropy is used as the loss function to calculate the comparison between the predicted emotional disorder labels and the true labels, and a graph regularization loss is added:
[0081]
[0082] The final loss function expression is:
[0083]
[0084] Among them, is the total loss function, which consists of two parts: the basic loss and the graph regularization loss λ is a hyperparameter that controls the weight of the graph regularization loss; in y i is the true class label of the i-th sample; is the model prediction probability of the i-th sample. In w ij is the weight matrix of the graph structure, representing the connection weight between nodes i and j; is the embedding of node i calculated by the graph neural network; is the embedding of node j calculated by the graph neural network; ||·|| 2 represents the square of the Euclidean norm, calculating the squared distance between two node embedding vectors.
[0085] S6: Use the optimal emotional disorder model to predict the emotional disorder state and conduct visual analysis.
[0086] The emotional disorder recognition system based on the asynchronous brain network includes
[0087] a data acquisition module for performing non-uniform EEG sampling, obtaining asynchronous EEG signals and resampling them;
[0088] A data processing module for preprocessing asynchronous electroencephalogram (EEG) signals to obtain aligned and normalized signal segments Among them, represents the k-th time segment data from the i-th EEG signal channel, where I is the total number of channels and K is the total number of time segments;
[0089] A first feature extraction module for extracting features from the aligned and normalized signal segments to obtain the asynchronous topological features of the signal segments;
[0090] A second feature extraction module for extracting features from the aligned and normalized signal segments to obtain the channel time-domain features of the signal segments;
[0091] A feature fusion module for converting the asynchronous topological features and channel time-domain features of the signal segments into a unified format and fusing them to obtain unified spatio-temporal fusion features;
[0092] A model construction module for constructing a preliminary emotional disorder recognition model, setting emotional disorder labels, inputting the spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and training the preliminary emotional disorder recognition model using an optimization strategy to obtain an optimal emotional disorder model;
[0093] An analysis module for predicting the emotional disorder state using the optimal emotional disorder model and performing visual analysis.
[0094] The first feature extraction module includes a first extraction unit and a first correlation analysis unit; the second feature extraction module includes a second extraction unit and a second correlation analysis unit.
[0095] Therefore, the present invention adopts the above-mentioned method and system for recognizing emotional disorders based on asynchronous brain networks, and through innovative theories and methods, develops a highly accurate, reliable and powerful computer-aided emotional disorder recognition model to significantly improve the accuracy of early judgment of emotional disorders.
[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. An emotional disorder recognition method based on an asynchronous brain network, characterized in that, It includes the following steps: S1: Perform non-uniform EEG sampling, obtain asynchronous EEG signals and resample them; S2: Perform primary preprocessing and secondary preprocessing on the asynchronous EEG signals to obtain the aligned and normalized signal segments wherein represents the data of the k-th time segment from the i-th EEG signal channel, I is the total number of channels, and K is the total number of time segments; the secondary preprocessing includes alignment and normalization processing S3: Extract the asynchronous topological features and channel time-domain features of the signal segments by performing feature extraction on the aligned and normalized signal segments; S4: Convert the asynchronous topological features and channel time-domain features of the signal segments into a unified format and fuse them to obtain unified spatio-temporal fusion features; S5: Construct a preliminary emotional disorder recognition model, set emotional disorder labels, input the spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and use an optimization strategy to train the preliminary emotional disorder recognition model to obtain the optimal emotional disorder model; S6: Use the optimal emotional disorder model to predict the emotional disorder state and perform visual analysis.
2. The method for identifying emotional disorders based on asynchronous brain networks according to claim 1, wherein The secondary preprocessing includes asynchronous alignment and normalization processing, specifically: Perform baseline drift correction on the asynchronous brain signals of each channel, use event trigger markers to perform time alignment on the asynchronous EEG signals of different channels, divide the aligned EEG signal data according to the event interval, perform time-domain alignment on the EEG signals within the event interval, intercept sequences of the same length, and perform amplitude normalization or Z-score normalization on the time-domain aligned EEG signals to obtain the aligned and normalized signal segments.
3. The method for identifying emotional disorders based on asynchronous brain networks according to claim 1, characterized in that The asynchronous topological features include: node degree, clustering coefficient, path length, betweenness centrality, eigenvector centrality, local efficiency, and global efficiency.
4. The method for identifying emotional disorders based on an asynchronous brain network according to claim 1, characterized in that Set a first feature extraction module to extract the asynchronous topological features of the EEG signals, including: Set a first extraction unit to perform feature extraction on the information segments to obtain local features; Set a global information acquisition module for acquiring the global structural information of the information segments; Set a first correlation analysis unit for predicting between local features according to the global structural information of the information segments, performing correlation analysis on the local features according to the prediction structure, setting a first correlation threshold, and removing the local features below the first correlation threshold to obtain the asynchronous topological features.
5. The method for identifying emotional disorders based on an asynchronous brain network according to claim 1, wherein Set a second feature extraction module to extract the channel time-domain features of the EEG signals, including: Set a second extraction unit to perform time-domain feature extraction on the information segments according to the information segments to obtain the first time-domain features; Perform position encoding on the first time-domain features, and set a second correlation analysis unit to analyze the correlation of the first time-domain features with position encoding, including the correlation of adjacent position information segments and the correlation of each information segment in the global information; Set a second correlation threshold, and remove the first time-domain features with position encoding below the second correlation threshold to obtain the second time-domain features, that is, the channel time-domain features.
6. The method for identifying emotional disorders based on asynchronous brain networks according to claim 1, wherein, The optimization strategy includes In the formula, Among them, is the total loss function, which consists of two parts: the base loss and the graph regularization loss λ is a hyperparameter that controls the weight of the graph regularization loss; in , y i is the true class label of the i-th sample; is the model prediction probability of the i-th sample. In , w ij is the weight matrix of the graph structure, representing the connection weight between nodes i and j; is the embedding of node i calculated by the graph neural network; is the embedding of node j calculated by the graph neural network; ||·|| 2 represents the square of the Euclidean norm, calculating the squared distance between two node embedding vectors.
7. An emotional disorder recognition system based on an asynchronous brain network, characterized in that, including A data acquisition module for performing non-uniform EEG sampling, obtaining asynchronous EEG signals and resampling them; A data processing module for preprocessing asynchronous EEG signals to obtain aligned and normalized signal segments Among them, represents the data of the k-th time segment from the i-th EEG signal channel, I is the total number of channels, and K is the total number of time segments; A first feature extraction module for performing feature extraction on the aligned and normalized signal segments to obtain the asynchronous topological features of the signal segments; A second feature extraction module for performing feature extraction on the aligned and normalized signal segments to obtain the channel time-domain features of the signal segments; A feature fusion module for converting the asynchronous topological features and channel time-domain features of the signal segments into a unified format and fusing them to obtain unified spatio-temporal fusion features; A model construction module, configured to construct a preliminary emotional disorder recognition model, set emotional disorder labels, input spatio-temporal fusion features and emotional disorder labels into the preliminary emotional disorder recognition model, and train the preliminary emotional disorder recognition model using an optimization strategy to obtain an optimal emotional disorder model; An analysis module, configured to predict an emotional disorder state using the optimal emotional disorder model and perform visual analysis.
8. The method for identifying emotional disorders based on an asynchronous brain network according to claim 7, wherein The first feature extraction module includes a first extraction unit and a first correlation analysis unit; the second feature extraction module includes a second extraction unit and a second correlation analysis unit.