Discovering sparse functional brain networks using group replicator dynamics (GRD).

Functional magnetic resonance imaging (fMRI) has become increasingly used for studying functional integration of the brain. However, the large inter-subject variability in functional connectivity renders detection of representative group networks very difficult. In this paper, we propose a new iterative method that we refer to as "group replicator dynamics," for detecting sparse functional networks that are common across subjects within a group. The proposed method uses replicator dynamics, which we show to be equivalent to non-negative sparse PCA, and incorporates group information for identifying common networks across subjects with subject-specific weightings of the identified brain regions reflecting individual differences. Finding a separate network for each subject, as opposed to employing traditional averaging approaches, permits statistical testing of group significance. We validated our method on synthetic data, and applying it to real fMRI data detected task-specific group networks that conform well with prior neuroscience knowledge.