A Survey of EEG Analysis based on Graph Neural Network

EEG signals reflect the activity of the brain. Previous studies based on EEG signal recognition focused on extracting temporal features of EEG and have not yet considered the topology of EEG signal channels, missing important spatial information. Graph neural networks view EEG signals as vertices of a graph and capture the implicit topological relationships between signals. In this review, we review the basic models of graph networks and their variants, including graph convolutional neural networks, graph attention networks, etc., summarize the applications in the fields of emotion recognition, epilepsy diagnosis, and finally present an outlook based on current challenges.