Linear Graph Convolutional Model for Diagnosing Brain Disorders

Abstract

Deep learning models find an increasing application in the diagnosis of brain disorders. Designed for large scale datasets, deep neural networks (DNNs) achieve state-of-the-art classification performance on a number of functional magnetic resonance imaging (fMRI) data. While utilizing DNNs might improve the performance, the complexity of the learning function decreases the interpretability of the model. Moreover, DNNs require considerably more time to train compared to their linear predecessors. In this paper, we re-examine the use of deep graph neural networks for graph-based disease prediction in favor of simpler linear models. We present a simplified linear model, which is more than 10 times faster to train than the previous DNN counterparts. We test our model on three fMRI datasets and show that it achieves comparable or superior performance to the state-of-the-art methods.