Integrating Structural and Functional Connectivity for Dynamic fMRI Modeling via Graph Diffusion Autoregression

Functional connectivity (FC) refers to the interregional associations between functional MRI signals from anatomically separate brain regions, providing critical insights into the brain's functional organization. Traditional FC approaches often treat connectivity as a stationary variable, relying solely on temporal cross-correlation while overlooking dynamic spatiotemporal patterns, and interaction with structural connectivity. These limitations restrict the sensitivity, specificity, and reliability of existing FC biomarkers. To overcome these limitations, we introduce the graph diffusion autoregressive (GDAR) model, a novel approach that integrates structural connectivity data from diffusion MRI into FC analysis and captures the dynamic, directional flow of communication signals across brain regions providing a more comprehensive assessment of the brain's connectome. Our results demonstrate that GDAR offers a reproducible measure of functional connectivity that differs fundamentally from traditional temporal cross-correlation analyses. It exhibits superior reproducibility compared to conventional fMRI analyses and is sensitive to physiological changes in the brain due to aging. These findings suggest that GDAR holds promise as an imaging biomarker for various neurological diseases.