Characterizing Brain Connectivity Alterations via Multimodal MRI Fusion for Alzheimer’s Disease

Exploring connectivity alterations in Alzheimer's diseases is important for understanding disease progression. Structural and functional MRI (magnetic resonance imaging), widely used in dementia assessment, provide complementary information, with sMRI capturing static anatomical structure and fMRI reflecting dynamic brain activity and functional connectivity. However, their different characteristics make it challenging to effectively integrate structural and functional information and consistently characterize disease-related connectivity changes. To address this challenge, we propose a multiscale, multimodal graph convolutional framework that jointly integrates structural and functional information across multiple spatial scales. At each scale, a graph convolutional network is dynamically constructed to adaptively model subject-specific connectivity patterns, enabling more flexible and expressive feature learning while preserving the spatial organization of each scale. The framework captures both coarse- and fine-grained connectivity patterns and provides interpretable representations of the relationships between structural changes and functional connectivity alterations. Experiments on the OASIS dataset demonstrate that the proposed multimodal framework improves diagnostic accuracy over single-modality and single-scale approaches while providing more informative and interpretable insights into disease-related brain alterations.