Learning High-Order Relationships with Hypergraph Attention-based Spatio-Temporal Aggregation for Brain Disease Analysis.

Abstract

Functional connectivity derived from functional magnetic resonance imaging (fMRI) primarily captures pairwise interactions between brain regions, which limits its ability to characterize complex high-order relationships. Hypergraph-based methods provide a natural way to model such interactions, yet most existing approaches rely on predefined hypergraph structures and neglect temporal dynamics, resulting in limited expressiveness and interpretability. To address these challenges, we propose a novel framework that jointly learns informative and sparse high-order brain structures along with their temporal dynamics. Inspired by the information bottleneck principle, we introduce an objective that maximizes information and minimizes redundancy, aiming to retain disease-relevant high-order features while suppressing irrelevant information. The proposed framework consists of three key components: (1) a multi-hyperedge binary mask module for hypergraph structure learning, (2) a hypergraph self-attention aggregation module that captures spatial features through adaptive attention across nodes and hyperedges, and (3) a spatio-temporal low-dimensional network for extracting discriminative spatio-temporal representations for disease classification. Experiments on benchmark fMRI datasets demonstrate that our method achieves competitive performance compared to the state-of-the-art approaches and effectively captures meaningful high-order brain interactions. These findings provide new insights into brain network modeling, showing potential for analyzing neuropsychiatric disorders.