Functional brain network classification method based on pre-training and graph neural network

Abstract

The invention discloses a functional brain network classification method based on pre-training and a graph neural network, and the method comprises the following steps: 1), obtaining fMRI data, and carrying out the preprocessing of the fMRI data; 2) performing brain region division and feature extraction on the fMRI data, and constructing a functional brain network in a graph form; 3) inputting the functional brain network without labels into a node coding layer for training; 4) aggregating network training through node information; 5) training the outputs of the step 3) and the step 4) through an edge relation prediction network; 6) inputting the functional brain network data with labels into the node coding layer trained in the step 3) for training; 7) performing training in the node information aggregation network trained in the step 4); and 8) performing training and classification through a functional brain network classification model. According to the invention, a large amount of label-free brain network data is utilized, and the graph neural network is pre-trained, so that the pre-trained network only needs to be trained on a small amount of label data to adapt to a downstream functional brain network classification task.

Description

technical field [0001] The present invention relates to the technical field of deep learning, in particular to a functional brain network classification method based on pre-training and graph neural network. Background technique [0002] Graph neural networks have powerful modeling capabilities in graph-structured data, but they also require high-quality labeled data for training. However, high-quality labeled data of brain networks are usually expensive and difficult to obtain, while correspondingly, a large amount of unlabeled data can be obtained relatively easily, such as based on open functional magnetic resonance imaging (fMRI) data such as the Human Connectome Project The set constructs functional brain network data, and its quantity can be thousands of times that of labeled data. Although the generation logic of network nodes and edges in various network data is different, as a graph structure, its inherent abstract features have something in common. If these unlabe...

AI Technical Summary

Problems solved by technology

However, high-quality labeled data of brain networks are usually expensive and difficult to obtain, while correspondingly, a large amount of unlabeled data can be obtained relatively easily, such ...

Images