A Method of Link Prediction of Sequential Functional Brain Networks Based on Generative Adversarial Network

Abstract

Purposes For the purpose of predicting the functional brain network and providing reference for studying the evolution patterns of functional brain network, a model of sequential brain function network based on Generative Adversarial Networks has been built. Methods The topological and temporal characteristics of brain function network are captured through Graph Convolutional Network and long-term and short-term memory network separately, and through feature fusion in the whole connection layer to realize the prediction of functional brain network. Findings The accuracy of network prediction with AUC and MAP indicators has been tested. The experimental results show that the AUC and MAP of the proposed method are 0.95 and 0.92, respectively on two different resting state fMRI data. Compared with other link prediction models, this method can achieve better prediction effect on functional brain network. The accurate prediction of brain function network owns a wide application prospect in the field of brain network decoding and brain computer interface.

Keywords

• generative adversarial network
• sequential link prediction
• graph convolutional network
• functional magnetic resonance