Multimetric brain functional-structural connectivity mutual coupling for mild cognitive impairment identification

Mild cognitive impairment (MCI) is widely recognized as a prodromal stage of many neurodegenerative conditions such as Alzheimer's disease, and its early identification is essential for timely clinical intervention. Multimodal MRI-derived functional connectivity (FC) and structural connectivity (SC) provide complementary views of brain network organization, and learning their coupling patterns has shown promising potential for MCI identification. However, most existing studies construct FC/SC networks using a single quantitative metric and model FC-SC relationships in a unidirectional manner, which may overlook heterogeneous topological characteristics and the intrinsic bidirectional interplay between brain function and structure. To cope with these, we propose a Multimetric Mutual Coupling Network (MMCN) for MCI identification. Specifically, multiple FC and SC metrics are employed to construct subject-specific multimetric connectivity graphs, enabling richer and more robust representations by mitigating single-metric bias. We further design an FC/SC-specific patch embedding module and introduce a cross-guided attention-based patch enhancement mechanism, which explicitly performs bidirectional feature refinement between FC and SC. This mutual coupling strategy facilitates knowledge-informed modeling of the interdependence between neuronal activity and white matter integrity, thereby capturing subtle yet critical connectivity disruptions associated with MCI. Extensive experiments on multimodal MRI data from one public dataset and one local cohort demonstrate the superior performance of MMCN comparing with state-of-the-art methods and highlight the necessity of multimetric-based FC-SC mutual coupling for accurate MCI identification. The codes of MMCN are publicly available at https://github.com/kevin-dgut/MMCN.