An evolving graph convolutional network for dynamic functional brain network

Brain networks have received extensive attention because of its important significance in understanding human brain organization and analyzing neuropsychiatric diseases. Existing methods are mostly based on the static functional brain network. However, the static brain network only considers the correlation between global signals, and cannot reflect the changes of brain information over time. In reality, the brain activities are constantly changing. Therefore, the dynamic functional brain network is proposed in order to reflect the variations of the signal. In recent years, as an important and effective method, graph convolutional network has been widely used in the analysis of brain networks. In consequence, an evolving graph convolutional network based on dynamic functional brain network is proposed. The network not only considers the neighbor node information in the current snapshot, but also considers the neighbor node information in the precursor and the subsequent time in the process of convolution. Furthermore, four kinds of convolution rules are put forward based on the evolving graph convolutional network. Alzheimer's disease diagnosis, as a representative neuropsychiatric diseases analysis method, is used to evaluate the model, and experiments are performed on the open dataset. The experimental results show that the proposed evolving graph convolutional network can improve the diagnostic accuracy to 99.16%, which proves the superiority of the proposed method.