In order to extend the ICA analysis from single-subject to multisubject study, the ICs estimated from each subject were clustered, matching the most similar spatial patterns across subjects. The self-organizing group ICA (sogICA) method, implemented in Brain Voyager QX, was used (Esposito et al., 2005). Once the ICs belonging to a cluster had been retrieved, the average spatial map was computed and assumed as representative for the cluster. The consistency of the clusters was expressed in terms by an intra-cluster similarity index s, defined as the average of the pair-wise spatial correlations between the constituting IC maps. The condition s>0.10 was used to select the clusters with reproducible maps across subjects. The remaining clusters were checked, analyzing also their single-subject maps. The exclusion of clusters that could be associated with artifacts was performed on the basis of the IC-fingerprint method (De Martino et al., 2007), as implemented in BrainVoyager QX.
