Topology-aware multi-view hypergraph computation-based cross-modal brain network fusion for brain disease diagnosis

Abstract

Cross-modal brain networks characterize the complex connections between different brain regions from both functional and structural perspectives. Deep fusion of functional and structural brain network information is crucial for brain disease diagnosis. However, existing methods overlook the intricate semantic and topological relationships between functional and structural brain networks, as well as their critical roles in brain network information transmission. To address these limitations, this paper proposes a topology-aware multi-view hypergraph computation-based cross-modal brain network fusion (TMHGC-CBNF) method. The TMHGC-CBNF achieves high-precision brain disease diagnosis through topology-aware multi-view brain network modeling, efficient message propagation, and multi-strategy fusion of cross-modal brain network information. Specifically, the topology-aware multi-view hypergraph computation method first constructs multi-view hypergraphs to model multi-level high-order correlations guided by topological structure and the semantic correlations under topological constraint in the functional brain network, while using graph structures to model the structural brain network. Based on this, parallel hypergraph convolutions are employed to simulate efficient information propagation patterns in the functional brain network and extract high-order feature representations for each view. Graph convolution is used to extract feature representations of the structural brain network. Next, a multi-strategy fusion method progressively and orthogonally fuses high-order functional brain network information from different views, and a topological encoding-based dual-channel cross-attention module facilitates the interaction of functional and structural brain network information in the topological space. Experiments on the three datasets demonstrate that the proposed method outperforms the current state-of-the-art methods and is capable of identifying cross-modal brain network biomarkers.