Developing a Graph Convolution-Based Analysis Pipeline for Multi-Modal Neuroimage Data: An Application to Parkinson's Disease


Parkinson’s disease (PD) is a highly prevalent neurodegenerative condition originating in subcortical areas of the brain and resulting in progressively worsening motor, cognitive, and psychiatric (e.g., depression) symptoms. Neuroimage data is an attractive research tool given the neurophysiological origins of the disease. Despite insights potentially available in magnetic resonance imaging (MRI) data, developing sound analytical techniques for this data has proven difficult. Principally, multiple image modalities are needed to compile the most accurate view possible; the process of incorporating multiple image modalities into a single holistic model is both poorly defined and extremely challenging. In this paper, we address these issues through the proposition of a novel graph-based convolutional neural network (GCN) architecture and present an end-to-end pipeline for preprocessing, formatting, and analyzing multimodal neuroimage data. We employ our pipeline on data downloaded from the Parkinson’s Progression Markers Initiative (PPMI) database. Our GCN model outperforms baseline models, and uniquely allows for direct interpretation of its results.