Subject neural feedback effectiveness prediction method and system based on graph neural network
By constructing resting-state fMRI brain maps using a graph neural network-based approach, and utilizing Pearson correlation coefficients and time-series statistics across brain regions, the problems of individual efficacy variability and insufficient model robustness in neurofeedback training were addressed, achieving efficient prediction of neurofeedback effectiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Chinese People's Liberation Army Cyberspace Force Information Engineering University
- Filing Date
- 2025-12-18
- Publication Date
- 2026-05-12
AI Technical Summary
Existing neurofeedback training suffers from individual variability in efficacy. Traditional methods lack robustness and generalization ability in predicting the effectiveness of neurofeedback, and feature extraction processes are complex and highly dependent.
A graph neural network-based approach was adopted, utilizing resting-state functional magnetic resonance imaging (fMRI) data to construct resting-state fMRI brain maps of subjects. Pearson correlation coefficients and time-series statistics of brain regions were extracted as node and edge features, and combined with graph convolutional networks to predict the effectiveness of neural feedback.
The brain image feature extraction process was simplified, the versatility and accuracy of the neural feedback effectiveness prediction model were improved, and the robustness and generalization ability of the prediction were enhanced.
Smart Images
Figure CN122004751A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent signal processing technology, and in particular to a method and system for predicting the effectiveness of subject neural feedback based on graph neural networks. Background Technology
[0002] Neurofeedback technology provides real-time feedback on brain activity to individuals, enabling them to autonomously control and manage their cognitive functions. It converts brain activity signals into feedback signals, allowing subjects to autonomously regulate their brain activity using specific strategies (such as autobiographical recall) based on the feedback information (e.g., thermometer readings). Neurofeedback training helps the brain learn self-regulation by monitoring brain signals in real-time (e.g., EEG, functional magnetic resonance imaging) and providing feedback, thereby improving functional status. Through real-time signal analysis, model adjustment, and feedback reinforcement, precise intervention can be achieved.
[0003] Functional magnetic resonance imaging (fMRI) utilizes the blood oxygen level dependent (BOLD) effect to compare and study the functional activity of the cerebral cortex, and has the advantage of high spatial resolution, making it one of the most popular experimental methods in neuroscience and neuroimaging. In 1995, Cox et al. proposed the concept of real-time functional magnetic resonance imaging (rtfMRI) to influence the behavior of subjects about to occur. In 2002, Yoo et al. first applied real-time fMRI to neurofeedback. This technology has been successfully applied in the treatment research of mental illnesses such as major depressive disorder, post-traumatic stress disorder, schizophrenia, and chronic insomnia. In recent years, traditional machine learning methods such as support vector machines and random forests based on resting-state brain indicators have often been used to predict the diagnostic and treatment effects of diseases. Graph Neural Networks (GNNs), as a neural network model specifically designed for processing graph-structured data, provide a new perspective for modeling complex data with their unique graph convolution and graph pooling mechanisms, further improving the model's versatility and robustness. In the problem of predicting the effectiveness of neurofeedback based on resting-state functional images of subjects, brain image feature extraction is a crucial element. Different types of predictive factors have a significant impact on the accuracy of prediction. After feature extraction, further screening or dimensionality reduction is usually required. Generally, important features are sorted and then put into the prediction model for classification, which makes the robustness and generalization ability of the neurofeedback effectiveness prediction model insufficient. Summary of the Invention
[0004] To address the issue of individual variability in the efficacy of neurofeedback training, this invention provides a method and system for predicting the effectiveness of neurofeedback in subjects based on graph neural networks. It utilizes time-series data from resting-state functional magnetic resonance imaging and a graph neural network model to predict the effectiveness of neurofeedback, simplifying the brain image feature extraction process and improving the versatility of the neurofeedback effectiveness prediction model.
[0005] According to the design scheme provided by the present invention, on the one hand, a method for predicting the effectiveness of subject neural feedback based on graph neural networks is provided, comprising:
[0006] Resting-state functional magnetic resonance imaging (fMRI) scans were performed on the subjects using a magnetic resonance scanner, and resting-state brain signal time series were extracted from the subjects. The brain signal time series recorded the changes in fMRI signal intensity of each brain region of interest at different time points.
[0007] Based on the resting-state brain signal time series, the Pearson correlation coefficient and corresponding brain region location index between the brain regions of interest of the subjects were obtained, and time series statistics were extracted by performing time-domain transformation on the brain signal time series.
[0008] Using time series statistics as node features and Pearson correlation coefficient and brain region location index to obtain edge features, a brain map based on the subject's resting-state fMRI is constructed.
[0009] Resting-state fMRI brain mapping data of the subjects were input into a pre-trained neural feedback effectiveness prediction model, which was then used to identify the effectiveness of the subjects' neural feedback.
[0010] As part of the present invention's method for predicting the effectiveness of subject neural feedback based on graph neural networks, the method further extracts the subject's resting-state brain signal time series, including:
[0011] The resting-state functional magnetic resonance imaging scan data are preprocessed, and the preprocessing includes: removing the first K time points, time-layer correction, head motion correction, structural image registration, spatial normalization, spatial smoothing and filtering, wherein K is greater than 5;
[0012] Based on the AAL template, multiple brain regions were selected as regions of interest to extract the time series of signals from the resting-state regions of interest of the subjects.
[0013] As part of the present invention's method for predicting the effectiveness of subject neural feedback based on graph neural networks, a further step is to construct a brain map based on the subject's resting-state fMRI, comprising:
[0014] Each brain region is taken as a node, and the time series statistics corresponding to the brain region are taken as node features. The time series statistics include: minimum value, maximum value, absolute maximum value, median, summation, variance, root mean square, standard deviation, mean and length of the time series signal.
[0015] The Pearson correlation coefficients between different brain regions were standardized, and the absolute values of the standardized Pearson correlation coefficients were used as edge weights between nodes. The node position information between edges was used as edge indexes to construct a brain map based on the subject's resting-state fMRI.
[0016] As a method for predicting the effectiveness of subject neural feedback based on graph neural networks according to the present invention, the neural feedback effectiveness prediction model further includes: an input layer for receiving resting-state fMRI brain map data of each subject, a hidden layer for extracting brain map features of the subject, and an output layer for label mapping output of brain map features. The hidden layer includes three convolutional blocks connected in sequence, a TopKPooling layer, a global max pooling layer, a global average pooling layer, and a fully connected layer. The first and second convolutional blocks use a GCN graph convolutional network layer to incorporate edge weights into edge features. The third convolutional block uses a GraphConv graph convolutional network layer to retain the original brain map information. The TopKPooling layer is set between the second and third convolutional layers to extract key node features in the brain map. The global max pooling layer and the global average pooling layer are used to perform global max pooling and global average pooling operations on the key node features and the original brain map features.
[0017] As a method for predicting the effectiveness of subject neural feedback based on graph neural networks in this invention, the global maximum pooling layer captures global features that describe the overall trend of brain map features by selecting the maximum value on each feature channel in the brain map features and combining it with the average value on each feature channel obtained by the global average pooling layer.
[0018] As a method for predicting the effectiveness of subject neural feedback based on graph neural networks in this invention, the outputs of the global max pooling layer and the global average pooling layer are further concatenated and fused in terms of dimension to form global features that are input to the fully connected layer for feature classification space mapping.
[0019] As a method for predicting the effectiveness of subject neural feedback based on graph neural networks in this invention, the fully connected layer further includes three sequentially connected linear transformation and regularization operations to map brain map features into a classification space through the three sequentially connected linear transformation and regularization operations.
[0020] Furthermore, this invention also provides a subject neural feedback effectiveness prediction system based on graph neural networks, comprising: a data acquisition module, a brain map modeling module, and a prediction output module, wherein,
[0021] The data acquisition module is used to perform resting-state functional magnetic resonance imaging (fMRI) scans on subjects using a magnetic resonance scanner and extract resting-state brain signal time series of subjects. The brain signal time series records the changes in fMRI signal intensity of each brain region of interest at different time points.
[0022] The brain map modeling module is used to obtain the Pearson correlation coefficient and corresponding brain region location index between brain regions of interest of the subject based on the resting-state brain signal time series, and to extract time series statistics by performing time-domain transformation on the brain signal time series; using the time series statistics as node features, and obtaining edge features based on the Pearson correlation coefficient and brain region location index, to construct a brain map based on the subject's resting-state fMRI.
[0023] The prediction output module is used to input the subject's resting-state fMRI brain map data into a pre-trained neural feedback effectiveness prediction model, and to use the neural feedback effectiveness prediction model to identify the subject's neural feedback effectiveness.
[0024] The beneficial effects of this invention are:
[0025] This invention is based on time-series data from resting-state functional magnetic resonance imaging (fMRI) of subjects before neurofeedback training. It uses functional connectivity between brain regions as edge features and time-series statistics of brain regions as point features. By constructing a graph neural network model, it can predict the effectiveness of neurofeedback, simplify the brain image feature extraction process, reduce the dependence of previous machine learning prediction models on features, and improve the versatility of the neurofeedback effectiveness prediction model. Attached image description:
[0026] Figure 1 This is a schematic diagram of the subject neural feedback effectiveness prediction process based on graph neural networks in the embodiment;
[0027] Figure 2 This is a schematic diagram of the neural feedback effectiveness prediction model structure in the embodiment. Detailed implementation method:
[0028] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and technical solutions.
[0029] The brain is not completely still at rest; spontaneous neural activity exists, forming functional networks. The state of these networks reflects the brain's current functional organization and is generally closely related to cognitive abilities and emotion regulation. Resting-state activity patterns in specific brain regions can predict an individual's response to neural feedback. Regarding the optimization of personalized intervention programs for neurofeedback training, this invention provides embodiments, see [link to embodiments]. Figure 1 As shown, a method for predicting the effectiveness of subject neural feedback based on graph neural networks is provided, which specifically includes the following:
[0030] S101. A resting-state functional magnetic resonance imaging (fMRI) scan was performed on the subject using a magnetic resonance scanner, and the resting-state brain signal time series of the subject was extracted. The brain signal time series recorded the changes in fMRI signal intensity of each brain region of interest at different time points.
[0031] The extraction of resting-state brain signal time series from subjects may include:
[0032] Preprocessing of resting-state functional magnetic resonance imaging scan data;
[0033] Based on the AAL template, multiple brain regions were selected as regions of interest to extract the time series of signals from the resting-state regions of interest of the subjects.
[0034] Data was acquired from subjects using an fMRI scanner. Subjects were instructed to remain awake and quiet during the resting-state fMRI scan. After acquisition, the fMRI data underwent preprocessing, including removal of the first 10 time points, time-to-layer correction, head motion correction (translation not exceeding 2.5 mm, rotation not exceeding 2.5°), structural image registration, spatial normalization, spatial smoothing (6 mm), and filtering (0.01-0.08 Hz). It should be noted that the number of time points removed during preprocessing can be adjusted according to the actual situation, such as removing the first 5 or 15 time points. The purpose is to eliminate noise interference caused by magnetic field instability during the early stages of scanner startup, thereby improving the signal-to-noise ratio of the data.
[0035] S102. Based on the resting-state brain signal time series, obtain the Pearson correlation coefficient and corresponding brain region location index between the brain regions of interest of the subjects, and extract time series statistics by performing time-domain transformation on the brain signal time series.
[0036] A total of 116 brain regions were identified based on the AAL template. Since the influence of the cerebellum was not considered, the first 90 brain regions were selected as brain regions of interest. Time series of resting-state brain signals of subjects before neurofeedback training were extracted based on these 90 brain regions. These time series recorded the changes in fMRI signal intensity of each brain region of interest at different time points.
[0037] Pearson correlation coefficients were calculated among 90 brain regions of the subjects to quantify the association between time series of two brain regions. Ten different statistical measures were extracted from the time series of each subject by performing time-domain transformation: minimum, maximum, absolute maximum, median, summation, variance, root mean square, standard deviation, mean, and length.
[0038] S103. Using time series statistics as node features, edge features are obtained based on Pearson correlation coefficient and brain region location index to construct a brain map based on the subject's resting-state fMRI.
[0039] Specifically, constructing a brain map based on the subject's resting-state fMRI can be designed to include:
[0040] Each brain region is taken as a node, and the time series statistics corresponding to the brain region are taken as node features. The time series statistics include: minimum value, maximum value, absolute maximum value, median, summation, variance, root mean square, standard deviation, mean and length of the time series signal.
[0041] The Pearson correlation coefficients between different brain regions were standardized, and the absolute values of the standardized Pearson correlation coefficients were used as edge weights between nodes. The node position information between edges was used as edge indexes to construct a brain map based on the subject's resting-state fMRI.
[0042] Considering that undirected graph modeling is commonly used in resting-state fMRI analysis, to ensure the integrity of the sample model while simplifying the model input and eliminating the interference of positive and negative correlations on model training, the absolute value of the standardized Pearson correlation coefficient between different brain regions was chosen as the edge weights for constructing the prediction model. Simultaneously, the node position information of each connection was preserved as the edge index, and the combination of these two factors served as the edge features of the prediction model. After time-domain transformation of the time series data, the minimum, maximum, absolute maximum, median, summation, variance, root mean square, standard deviation, mean, and length were extracted as the node features of the prediction model.
[0043] S104. Input the subject's resting-state fMRI brain map data into a pre-trained neural feedback effectiveness prediction model, and use the neural feedback effectiveness prediction model to identify the subject's neural feedback effectiveness.
[0044] The neural feedback effectiveness prediction model can be designed as follows: an input layer for receiving resting-state fMRI brain map data of each subject, a hidden layer for extracting brain map features of the subject, and an output layer for label mapping of brain map features. The hidden layer includes three convolutional blocks connected in sequence, a TopKPooling layer, a global max pooling layer, a global average pooling layer, and a fully connected layer. The first and second convolutional blocks use a GCN graph convolutional network layer to incorporate edge weights into edge features. The third convolutional block uses a GraphConv graph convolutional network layer to retain the original brain map information. The TopKPooling layer is set between the second and third convolutional layers to extract key node features in the brain map. The global max pooling layer and the global average pooling layer are used to perform global max pooling and global average pooling operations on the key node features and the original brain map features.
[0045] The global max pooling layer captures global features describing the overall trend of the brain map features by combining the maximum value of each feature channel obtained from the global average pooling layer with the average value of each feature channel. The outputs of the global max pooling layer and the global average pooling layer are concatenated and fused in dimensionality to form global features that are input to the fully connected layer for feature classification space mapping. The fully connected layer includes three sequentially connected linear transformation and regularization operations to map the brain map features to the classification space.
[0046] like Figure 2As shown, the resting-state fMRI brain map of each subject is input into the model in the input layer. In the output layer, the obtained global features are mapped to the label dimension to achieve effective group identification. The label value of the effective group is set to 1, and the label value of the invalid group is set to 0. The hidden layer design mainly consists of three convolutional blocks, with an independent BN layer added after each convolutional block to accelerate convergence, alleviate gradient problems, and enhance the model's nonlinear expressive ability. In the first and second convolutional blocks, a Graph Convolutional Network (GCN) that can consider edge weights is designed, allowing the model to incorporate edge weights into the edge features. The first convolutional block maps the input node features to a 128-dimensional latent space. Then, the 128-dimensional node features are processed by normalized batch processing and the ReLU activation function to accelerate training and improve the model's stability. The second convolutional block has a similar structure to the first convolutional block, with an input-output dimension of 128, to further extract feature information. A TopKPooling pooling layer is designed between the two graph convolutional layers to further extract key node information from the mind map data. A GraphConv layer, also a graph convolutional layer with 128 input and output dimensions, is added after the two GCN layers. To effectively capture global features after the graph pooling layer, a specific method is employed: after the second and third convolutional layers, global max pooling (GMP) and global average pooling (GAP) are used to generate global features, which are then concatenated to create low-level and high-level features. GMP features capture salient features in the graph by selecting the maximum value in each feature channel, while GAP features capture the overall trend by calculating the average value of each feature channel. By concatenating the 128-dimensional features of GMP and GAP, a 256-dimensional global feature is formed, creating a complementary description of the features in this layer.
[0047] Before entering the fully connected layer, neurons are first randomly masked with a probability of 0.2 using a Dropout layer. Then, after three layers of linear transformations and regularization, the model is finally mapped to the classification space. For each linear transformation, sparsity is introduced using the ReLU activation function: the first layer maps 256-dimensional features to 128-dimensional features, the second layer maps 128-dimensional features to 32-dimensional features, and the third layer maps 32-dimensional features to the number of classes. A Dropout operation is added after the first two linear transformations to further reduce the risk of overfitting. Finally, the output is classified using Log-Softmax.
[0048] After training a predictive model using machine learning methods, the model can predict an individual's response to neural feedback based on resting-state characteristics. Resting-state brain mapping data of new subjects are input into the model to output the predicted effects of their neurofeedback therapy, thereby guiding personalized interventions.
[0049] Furthermore, based on the above method, this embodiment of the invention also provides a subject neural feedback effectiveness prediction system based on graph neural networks, comprising: a data acquisition module, a brain map modeling module, and a prediction output module, wherein,
[0050] The data acquisition module is used to perform resting-state functional magnetic resonance imaging (fMRI) scans on subjects using a magnetic resonance scanner and extract resting-state brain signal time series of subjects. The brain signal time series records the changes in fMRI signal intensity of each brain region of interest at different time points.
[0051] The brain map modeling module is used to obtain the Pearson correlation coefficient and corresponding brain region location index between brain regions of interest of the subject based on the resting-state brain signal time series, and to extract time series statistics by performing time-domain transformation on the brain signal time series; using the time series statistics as node features, and obtaining edge features based on the Pearson correlation coefficient and brain region location index, to construct a brain map based on the subject's resting-state fMRI.
[0052] The prediction output module is used to input the subject's resting-state fMRI brain map data into a pre-trained neural feedback effectiveness prediction model, and to use the neural feedback effectiveness prediction model to identify the subject's neural feedback effectiveness.
[0053] To verify the effectiveness of this solution, the following explanation is based on experimental data:
[0054] The experiment was validated using a real-time functional magnetic resonance imaging (fMRI) neurofeedback training dataset from 51 subjects. The scales before and after neurofeedback training were used as the screening criteria, with the label value set to 1 for the 31 subjects in the effective neurofeedback group and 0 for the 20 subjects in the ineffective group. Resting-state functional connectivity before neurofeedback training was calculated as edge features based on 90 brain regions using the AAL template, and 10-dimensional time-series statistics corresponding to the 90 brain regions were used as node features. A five-fold cross-validation method was employed, dividing the dataset into five subsets. Four subsets were used as training data each time, and the remaining subset was used as validation data, repeated five times. To minimize randomness and ensure stable results, 10 experiments were conducted with different seed numbers. The average prediction accuracy was 81.42%, achieving excellent classification performance.
[0055] Experimental data show that the proposed solution, based on time-series data from resting-state functional magnetic resonance imaging of subjects before neurofeedback training, using functional connectivity between brain regions as edge features and time-series statistics of brain regions as point features, can effectively predict the effectiveness of neurofeedback through a graph neural network model.
[0056] Unless otherwise specifically stated, the relative steps, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention.
[0057] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0058] The units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations are not considered to be beyond the scope of this invention.
[0059] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware or as a software functional module. This invention is not limited to any particular combination of hardware and software.
[0060] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for predicting the effectiveness of subject neural feedback based on graph neural networks, characterized in that, Include: Resting-state functional magnetic resonance imaging (fMRI) scans were performed on the subjects using a magnetic resonance scanner, and resting-state brain signal time series were extracted from the subjects. The brain signal time series recorded the changes in the fMRI signal intensity of each brain region of interest at different time points. Based on the resting-state brain signal time series, the Pearson correlation coefficient and corresponding brain region location index between the brain regions of interest of the subjects were obtained, and time series statistics were extracted by performing time-domain transformation on the brain signal time series. Using time series statistics as node features and Pearson correlation coefficient and brain region location index to obtain edge features, a brain map based on the subject's resting-state functional magnetic resonance imaging is constructed. Resting-state functional magnetic resonance imaging (fMRI) brain mapping data of the subjects were input into a pre-trained neural feedback effectiveness prediction model, which was then used to identify the effectiveness of the subjects' neural feedback.
2. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 1, characterized in that, Extracting time series of resting-state brain signals from subjects, including: The resting-state functional magnetic resonance imaging scan data are preprocessed, and the preprocessing includes: removing the first K time points, time-layer correction, head motion correction, structural image registration, spatial normalization, spatial smoothing and filtering, wherein K is greater than 5; Based on the AAL template, multiple brain regions were selected as regions of interest to extract the time series of signals from the resting-state regions of interest of the subjects.
3. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 1, characterized in that, Constructing a brain map based on the subject's resting-state functional magnetic resonance imaging (fMRI), including: Each brain region is taken as a node, and the time series statistics corresponding to the brain region are taken as node features. The time series statistics include: minimum value, maximum value, absolute maximum value, median, summation, variance, root mean square, standard deviation, mean and length of the time series signal. The Pearson correlation coefficients between different brain regions were standardized, and the absolute values of the standardized Pearson correlation coefficients were used as edge weights between nodes. The node position information between edges was used as edge indexes to construct a brain map based on the subject's resting-state functional magnetic resonance imaging.
4. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 1, characterized in that, The neural feedback effectiveness prediction model includes: an input layer for receiving resting-state functional magnetic resonance imaging (fMRI) brain map data from each subject; a hidden layer for extracting brain map features from the subject; and an output layer for label mapping of the extracted features. The hidden layer comprises three sequentially connected convolutional blocks, a TopKPooling layer, a global max pooling layer, a global average pooling layer, and a fully connected layer. The first and second convolutional blocks use a GCN graph convolutional network layer to incorporate edge weights into the edge features. The third convolutional block uses a GraphConv graph convolutional network layer to retain the original brain map information. The TopKPooling layer is positioned between the second and third convolutional layers to extract key node features from the brain map. The global max pooling layer and the global average pooling layer perform global max pooling and global average pooling operations on the key node features and the original brain map features, respectively.
5. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 4, characterized in that, The global max pooling layer captures global features that describe the overall trend of the brain map structure features by combining the maximum value on each feature channel of the brain map features with the average value on each feature channel obtained by the global average pooling layer.
6. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 4 or 5, characterized in that, The outputs of the global max pooling layer and the global average pooling layer are concatenated and fused in the dimension to form global features that are input to the nonlinear transformation layer for feature classification space mapping.
7. The method for predicting the effectiveness of subject neural feedback based on graph neural networks according to claim 4, characterized in that, The fully connected layer includes three sequentially connected linear transformation and regularization operations to map brain map features to a classification space.
8. A subject neural feedback effectiveness prediction system based on graph neural networks, characterized in that, It includes: a data acquisition module, a mind map modeling module, and a prediction output module, among which, The data acquisition module is used to perform resting-state functional magnetic resonance imaging (fMRI) scans on subjects using a magnetic resonance scanner and extract resting-state brain signal time series of subjects. The brain signal time series records the changes in the fMRI signal intensity of each brain region of interest at different time points. The brain mapping modeling module is used to obtain the Pearson correlation coefficient and corresponding brain region location index between brain regions of interest in the subject based on the resting-state brain signal time series, and to extract time series statistics by performing time-domain transformation on the brain signal time series; using the time series statistics as node features, and obtaining edge features based on the Pearson correlation coefficient and brain region location index, to construct a brain map based on the subject's resting-state functional magnetic resonance imaging. The prediction output module is used to input the subject's resting-state functional magnetic resonance brain map data into a pre-trained neural feedback effectiveness prediction model, and to use the neural feedback effectiveness prediction model to identify the effectiveness of the subject's neural feedback.
9. An electronic device, characterized in that, include: At least one processor, and a memory coupled to said at least one processor; The memory stores a computer program that can be executed by the at least one processor to implement the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, enables the implementation of the method as described in any one of claims 1 to 7.