Extracting Connectomic Profiles from Group Resting State fMRI Data Using Dictionary Learning

Kaiming LI¹, Xiaoping P. Hu¹

This paper describes a new framework to characterize the connectomic profiles for distinct functional regions on the cortical surface. Unlike existing group ICA approaches that heavily rely on spatial smoothing and registration techniques, this framework employs two measures, cortical parcellation by BOLD signal homogeneity and over complete dictionary learning, to account for the well-known anatomical variability across individuals. Our results show that the resultant connectomic profiles are robust and can be used for the identification of both distinct functional regions and functional networks, facilitating building statistical models for these profiles and pinpointing disrupted regions in pathological/psychiatric brain disorder datasets.