Inferring Human Brain Structural Connectivity Based on Neural Networks

Abstract

A central and fundamental issue in cognitive neuroscience is to comprehend the relationship between human brain functional and structural connectivity. Previous studies normally focus on the relationship by predicting functional connectivity from structural connectivity and show there is a cohesive correlation between the two types of networks. In this paper, we investigate the relation by revealing the true anatomical connections from the functional correlations using multi-layer neural networks, which is trained to learn the intrinsic mapping mechanism and recover some missed connections with diffusion magnetic resonance imaging (dMRI) tractography, particularly the cross-hemispheric homotopic connections. We execute the method to a dataset with 246 brain areas acquired from 147 subjects. The results show that around 65% of the average intrahemispheric structural connections are correctly inferred.