CN-122004751-A - Method and system for predicting subject neural feedback effectiveness based on graph neural network

CN122004751ACN 122004751 ACN122004751 ACN 122004751ACN-122004751-A

Abstract

The invention relates to the technical field of signal intelligent processing, in particular to a method and a system for predicting the neural feedback effectiveness of a subject based on a graph neural network, which adopt a magnetic resonance scanner to perform resting state functional magnetic resonance imaging scanning on the subject and extract a resting state brain signal time sequence of the subject; the method comprises the steps of obtaining pearson correlation coefficients and corresponding brain region position indexes among brain regions of interest of a subject based on a resting brain signal time sequence, extracting time sequence statistics through time domain transformation of the brain signal time sequence, obtaining edge characteristics according to the pearson correlation coefficients and the brain region position indexes by taking the time sequence statistics as node characteristics, constructing a brain map based on resting fMRI of the subject, and identifying the nerve feedback effectiveness of the subject by using a nerve feedback effectiveness prediction model. The invention can simplify the brain image feature extraction process, reduce the feature dependence degree of the traditional machine learning prediction model construction, and promote the universality of the nerve feedback effectiveness prediction model.

Inventors

YAN BIN
WANG LINYUAN
LI ZHONGLIN
LIU AO
GAO HUI
WU HAOWEI
LI YONGLI
ZOU ZHI
SUN YONGBING
CAO LONG
LI LEI
ZHANG CHI

Assignees

中国人民解放军网络空间部队信息工程大学

Dates

Publication Date: 20260512
Application Date: 20251218

Claims (10)

1. A method for predicting the effectiveness of neural feedback of a subject based on a graph neural network, comprising: performing resting state functional magnetic resonance imaging scanning on a subject by adopting a magnetic resonance scanner, and extracting a resting state brain signal time sequence of the subject, wherein the brain signal time sequence records the functional magnetic resonance signal intensity change of each brain region of interest at different time points; Acquiring pearson correlation coefficients and corresponding brain region position indexes between brain regions of interest of a subject based on a resting brain signal time sequence, and extracting time sequence statistics by performing time domain transformation on the brain signal time sequence; Taking the time sequence statistics as node characteristics, acquiring edge characteristics according to the Pearson correlation coefficient and the brain region position index, and constructing a brain map based on the resting state function magnetic resonance of the subject; and inputting brain map data of the resting state functional magnetic resonance of the subject into a pre-trained nerve feedback effectiveness prediction model, and identifying the nerve feedback effectiveness of the subject by using the nerve feedback effectiveness prediction model.
2. The method for predicting the effectiveness of neural feedback in a subject based on a graph neural network of claim 1, wherein extracting a time series of resting brain signals in the subject comprises: Preprocessing resting state functional magnetic resonance imaging scan data, wherein the preprocessing comprises removing the previous K time points, time layer correction, head motion correction, structural image registration, spatial standardization, spatial smoothing and filtering, wherein K is more than 5; based on brain region division of the AAL template, selecting a plurality of brain regions as brain regions of interest to extract a resting state brain region signal time sequence of interest of the subject.
3. The method for predicting the effectiveness of neural feedback in a subject based on a graph neural network of claim 1, wherein constructing a brain graph based on resting state functional magnetic resonance of the subject comprises: Taking each brain area as a node, and taking time sequence statistics corresponding to the brain area as node characteristics, wherein the time sequence statistics comprise the minimum value, the maximum value, the absolute maximum value, the median, the summation, the variance, the root mean square, the standard deviation, the mean value and the length of time sequence signals; And normalizing the pearson correlation coefficients among different brain areas, taking the absolute value of the normalized pearson correlation coefficients as the edge weight among the nodes, and taking the node position information among the edges as the edge index to construct a brain map based on the resting state functional magnetic resonance of the subject.
4. The method for predicting the neural feedback effectiveness of a subject based on a graph neural network according to claim 1, wherein the neural feedback effectiveness prediction model comprises an input layer for receiving brain graph data of each subject resting state functional magnetic resonance, a hidden layer for extracting brain graph characteristics of the subject, and an output layer for performing label mapping output on the extracted characteristics, wherein the hidden layer comprises three layers of convolution blocks and TopKPooling layers, a global maximum pooling layer, a global average pooling layer and a full connection layer which are sequentially connected, the first layer of convolution blocks and the second layer of convolution blocks adopt a GCN graph convolution network layer to blend edge weights into edge characteristics, the third layer of convolution blocks adopts a GraphConv graph convolution network layer to reserve original brain graph information, the TopKPooling layers are arranged between the second convolution layer and the third convolution layer to extract key node characteristics in the brain graph, and the global maximum pooling layer and the global average pooling layer are used for performing global maximum pooling operation and global average pooling operation on the key node characteristics and the original brain graph characteristics.
5. The method for predicting the effectiveness of neural network based on subject neural feedback of claim 4, wherein the global maximization layer captures global features describing overall trends of the structural features of the brain map by selecting a maximum value on each of the feature channels in the features of the brain map in combination with an average value on each of the feature channels obtained by the global averaging pooling layer.
6. The method for predicting the effectiveness of neural feedback of a subject based on a graph neural network of claim 4 or 5, wherein the outputs of both the global max pooling layer and the global average pooling layer are feature stitched and fused in dimensions to form global features that are input to the nonlinear transformation layer for feature classification space mapping.
7. The graph neural network-based subject neural feedback effectiveness prediction method of claim 4, wherein the fully connected layer includes three layers of sequentially connected linear transformation and regularization operations to map brain graph features into the classification space through the three layers of sequentially connected linear transformation and regularization operations.
8. A subject nerve feedback effectiveness prediction system based on a graph nerve network is characterized by comprising a data acquisition module, a brain graph modeling module and a prediction output module, wherein, The data acquisition module is used for carrying out resting state functional magnetic resonance imaging scanning on a subject by adopting a magnetic resonance scanner and extracting a resting state brain signal time sequence of the subject, wherein the brain signal time sequence records the functional magnetic resonance signal intensity change of each brain region of interest at different time points; the brain map modeling module is used for acquiring the Pearson correlation coefficient and the corresponding brain region position index between brain regions of interest of a subject based on the resting state brain signal time sequence, and extracting time sequence statistics by performing time domain transformation on the brain signal time sequence; and the prediction output module is used for inputting the resting state functional magnetic resonance brain map data of the subject to a pre-trained nerve feedback effectiveness prediction model, and identifying the nerve feedback effectiveness of the subject by using the nerve feedback effectiveness prediction model.
9. An electronic device, comprising: At least one processor, and a memory coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor to implement the method of any one of claims 1-7.
10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, which, when executed, is capable of implementing the method according to any one of claims 1-7.

Description

Method and system for predicting subject neural feedback effectiveness based on graph neural network Technical Field The invention relates to the technical field of signal intelligent processing, in particular to a method and a system for predicting the effectiveness of nerve feedback of a subject based on a graph neural network. Background The nerve feedback technology is to feed back brain activities to an individual in real time so as to achieve autonomous control and management of brain cognitive functions by the individual. It converts the brain activity signal into a form of feedback signal, which allows the subject to autonomously regulate brain activity in a certain regulation strategy (self-memory, etc.) based on the feedback information (thermometer, etc.). Neural feedback training helps brain learn self-regulation by monitoring brain signals (such as brain electricity, functional magnetic resonance and the like) in real time and providing feedback, so that functional states are improved, and accurate intervention is realized through real-time signal analysis, model adjustment and feedback reinforcement. Functional magnetic resonance imaging (functional Magnetic Resonance Imaging, fMRI) uses blood oxygen level dependent (blood oxygen LEVEL DEPENDENT, BOLD) effects to contrast and study the functional activities of cerebral cortex, and has the advantage of high spatial resolution, which is one of the most popular experimental methods in the fields of neuroscience and neuroimaging. Cox et al in 1995 proposed the concept of real-time functional magnetic resonance imaging (real-time functional Magnetic Resonance Imaging, rtfMRI) to affect the impending behavior of the subject. Yoo et al in 2002 applied a real-time functional magnetic resonance imaging technique to nerve feedback for the first time. The technology has been successfully applied to the treatment and research of mental diseases such as major depressive disorder, post-traumatic stress disorder, schizophrenia, chronic insomnia disorder and the like. In recent years, traditional machine learning methods such as a support vector machine and a random forest based on brain resting state indexes are often used for predicting diagnosis and treatment effects of diseases. The graph neural network (Graph Neural Network, GNN) is used as a neural network model specially processing graph structure data, and by means of a unique graph rolling and pooling mechanism, a new view angle is provided for complex data modeling, and the universality and the robustness of the model are further improved. In the problem of predicting the effectiveness of the nerve feedback aiming at the resting state functional image of the subject, the extraction of brain image features is a key element, different types of predictors have great influence on the accuracy of prediction, the features usually need to be further screened or reduced in dimension after the extraction, and important features are generally obtained according to the sequence and then are put into a prediction model for classification, so that the robustness and generalization capability of the nerve feedback effectiveness prediction model are insufficient. Disclosure of Invention Aiming at the problem of individual curative effect variability in nerve feedback training, the invention provides a method and a system for predicting the effectiveness of subject nerve feedback based on a graph neural network, and the prediction of the effectiveness of the nerve feedback is realized by utilizing the time sequence data of the resting-state functional magnetic resonance imaging and based on a graph neural network model, the brain image feature extraction flow is simplified, and the universality of the prediction model of the effectiveness of the nerve feedback is improved. According to an aspect of the design scheme provided by the invention, a method for predicting the effectiveness of nerve feedback of a subject based on a graph neural network is provided, which comprises the following steps: Performing resting state functional magnetic resonance imaging scanning on a subject by adopting a magnetic resonance scanner, and extracting a resting state brain signal time sequence of the subject, wherein the brain signal time sequence records fMRI signal intensity changes of each brain region of interest at different time points; Acquiring pearson correlation coefficients and corresponding brain region position indexes between brain regions of interest of a subject based on a resting brain signal time sequence, and extracting time sequence statistics by performing time domain transformation on the brain signal time sequence; Taking the time sequence statistic as a node characteristic, acquiring an edge characteristic according to the Pearson correlation coefficient and the brain region position index, and constructing a brain map based on resting state fMRI of the subject; Inputting the resting fMRI brain map data of th