Dynamic Fusion of Structural and Functional Connectivity via Joint Connectivity-Matrix ICA

Integrating functional MRI (fMRI) and diffusion MRI (dMRI) advances neuroimaging by merging complementary perspectives on brain networks, with fMRI measuring functional activity and dMRI capturing structural connectivity (SC). However, combining them dynamically remains challenging due to their drastically different data characteristics.

We propose a novel framework, "dynamic fusion," which extends joint component analysis via connectivity-matrix ICA to integrate static SC with dynamic functional connectivity (FC). It evaluates how joint components relate across temporal states to capture both static and time-varying connectivity features. As a proof-of-concept, and in contrast to more systematic approaches for modeling true temporal dynamics, we defined two temporal states from the first and last thirds of the fMRI time series. We applied this to fMRI and dMRI from control subjects previously analyzed with a static-only model and to a comparable schizophrenia group from the same study.

Results revealed heterogeneous temporal dynamics in FC and, importantly, showed that SC representations differ across the two functional time segments, reflecting functional-context-dependent decomposition rather than anatomical change. It also detected joint dynamic SC-FC differences between schizophrenia and control groups, indicating sensitivity to group-level effects that may be missed in static analyses.

Although the current time segments are not intended to represent canonical or recurring brain states, the results suggest that multimodal fusion outcomes can depend on the functional context with which SC is fused. Overall, our dynamic fusion offers a promising approach for integrating SC and dynamic FC to better understand brain organization and neuropsychiatric disorders, while motivating more systematic definitions of dynamic states in future work.