Neurobridge: Bridging functional and structural brain networks via neural coupling and consistency-Guided dynamic graph learning.

Abstract

Modern medical imaging provides important insights into brain network analysis. Functional brain networks are used to characterize the functional connectivity patterns in resting or task states, and structural brain networks reflect the integrity and connectivity strength of macro-scale pathways. However, there are differences in data structure, information representation and spatial resolution between the both, and how to effectively fuse the information from these two modalities to mine potential cross-modal representations has become a key challenge in current research. In this paper, we propose the NeuroBridge, which enable the interaction between different modalities and the extraction of discriminative joint representations through coupling at the macro-scale level for brain network analysis. Specifically, the Neural Synergy Coupling Module (NeuSCM) performs structural-functional coupling in terms of brain region receptive fields. In order to enhance the inter-modal high-level semantic spatial coherence, we propose the Consistency Anchor Guidance Module (CAGM) for semantic calibration and convergence control of the fused representation space. Finally the Dynamic Association Parsing Module (DAP) captures the complex relationships between nodes and is used for final prediction and biomarker extraction. We conduct extensive experiments on disease prediction and gender classification tasks, and our results show that the prediction accuracy of our method outperforms that of SOTA single and multimodal methods, while our study provides new insights into multimodal brain network analysis.