1. INTRODUCTION
Neuroimaging pattern classification methods have demonstrated recent advances in predicting Alzheimer’s disease (AD) and mild cognitive impairment (MCI) from magnetic resonance imaging (MRI) and positron emission tomography (PET) scans [1], [2], [3]. Since the brain is an extremely complex system, large improvements in understanding the brain’s organization have been made by representing the brain as a connectivity graph [4]. In this graph, nodes are defined as brain regions of interest (ROIs) and edges are defined as the connectivity between those ROIs. This representation is highly compatible with Graph Convolutional Network (GCN), a deep learning method with demonstrated capabilities for analyzing graph structure problems [5], [6], [7], [8].