Group information guided ICA for fMRI data analysis.

Group independent component analysis (ICA) has been widely applied to studies of multi-subject fMRI data for computing subject specific independent components with correspondence across subjects. However, the independence of subject specific independent components (ICs) derived from group ICA has not been explicitly optimized in existing group ICA methods. In order to preserve independence of ICs at the subject level and simultaneously establish correspondence of ICs across subjects, we present a new framework for obtaining subject specific ICs, which we coined group-information guided ICA (GIG-ICA). In this framework, group information captured by standard ICA on the group level is exploited as guidance to compute individual subject specific ICs using a multi-objective optimization strategy. Specifically, we propose a framework with two stages: at first, group ICs (GICs) are obtained using standard group ICA tools, and then the GICs are used as references in a new one-unit ICA with spatial reference (ICA-R) using a multi-objective optimization solver. Comparison experiments with back-reconstruction (GICA1 and GICA3) and dual regression on simulated and real fMRI data have demonstrated that GIG-ICA is able to obtain subject specific ICs with stronger independence and better spatial correspondence across different subjects in addition to higher spatial and temporal accuracy.