Classification of Alzheimer’s Disease via Spatial-Temporal Graph Convolutional Networks

In recent years, functional brain networks have been widely used for brain disease classification tasks. Graph convolutional networks (GCN), as one of the existing methods, focuses on inter-subject relationships and subjects’ brain region connectivity relationships. Though spatial features can be extracted by GCN-based methods from functional brain network, temporal relationships between brain regions are frequently ignored. In this paper, we propose a spatial-temporal graph convolutional network (STGCN) model to solve this problem. A temporal convolution module (TCM) is designed to learn information about the alteration of brain regions with time, and a spatial convolution module (SCM) is employed to extracted features of brain regions. TCM contains two parts, single-brain-region convolution and multi-brain-region convolution, to fully capture the alteration of brain regions with time from two perspectives. We finally evaluate our model on the ADNI dataset and obtain good results.