Attend to Connect: End-to-End Brain Functional Connectivity Estimation

Abstract

Functional connectivity (FC) studies have demonstrated the benefits of investigating the brain and its disorders through the undirected weighted graph of fMRI correlation matrix. Most of the work with the FC, however, depends on the way the connectivity is computed and further depends on the manual post-hoc analysis of the FC matrices. In this work, we propose a deep learning architecture (BrainGNN) that learns the connectivity structure while learning to classify subjects. It simultaneously trains a graphical neural network on this graph and learns to select a sparse subset of brain regions important to the prediction task. We demonstrate the model's state-of-the-art classification performance on a schizophrenia fMRI dataset and show how introspection leads to disorder-relevant findings. The graphs learned by the model exhibit strong class discrimination, and the identified sparse subset of relevant regions is consistent with the schizophrenia literature.