Structure-Function Coupling in the Human Connectome with Hypergraph Neural Networks

Structure-function coupling in the human connectome is a critical research area in neuroscience. Understanding the relationship between the brain’s anatomical architecture and its functional dynamics is essential for exploring brain function and developing treatments for neurological disorders. However, traditional methods, including linear models and computational techniques, have made some progress but struggle to capture the complex nonlinear relationships within brain networks. Here, we propose a Hypergraph Neural Network (HGNN) framework to map the structural connectome to individual brain function. Specifically, we utilize hypergraphs to capture the intricate multi-way relationships and interactions between brain regions. It utilizes graph convolution and graph attention operators to update hyperedge weights, effectively mapping them to functional connectivity (FC). Our approach significantly enhances prediction accuracy compared to other methods, as evidenced by the Human Connectome Project (HCP) dataset. The results indicate that the structure-function coupling in the human brain network strengthens at higher dimensions and is primarily dominated by group-common factors. These findings underscore the HGNN model’s potential for unraveling the complex links between brain structure and function.