BCN-GCN: A Novel Brain Connectivity Network Classification Method via Graph Convolution Neural Network for Alzheimer’s Disease

Brain connectivity network (BCN) is a non-directed graph which represents the relationship between brain regions. Traditional feature extractor methods has been applied to identify diseases related to brains such as Alzheimer’s disease (AD) with BCN. However, the feature extraction capabilities of these methods are insufficient to collect various topological properties of non-directed graphs. They can only take into consider the local topological properties or frequent sub-graph topological properties. To explicitly exploit these properties, this paper proposes a BCN based fully-supervised Graph Convolution Network (GCN) to select features and classify disease automatically, skipping the step of manual feature selection and keeping all information in the brain connectivity network. Extensive experiments show that we outperform other method in classifying different stages of AD. Furthermore, our method can find the most relevant brain regions to classify different stages of AD, showing better interpretation ability compared to traditional methods.