Spatio-Temporal Model for Dynamic Functional Connectivity in Resting State fMRI Analysis

Functional connectivity (FC) in fMRI exhibits spatiotemporal dynamics. A new fine-grained, single subject modeling of this dynamic FC is introduced based on functional connectivity units (FCUs), defined as a small set of anatomical brain regions. The estimation of FCUs is considered as a dictionary learning problem which is solved with an ADMM algorithm. A complex synthetic data generation procedure is developed to mimic brain dynamics and proved the performance of the proposed algorithm. The analyses of resting state fMRI signals of mice revealed that both the spatial structure and the temporal activity of the obtained FCUs are biologically valid and show high inter-subject coherence.