I have run a group ICA on resting state data after manual cleaning using MELODIC.

We have found good results for the comparisons of some ICs but others give large, unspecific differences between the groups. 

I have tried to look through the dr_stage2_ic#### files for obvious outliers, but the pattern of spikes vary a lot when checking different voxels, even after demeaning the whole set.

I have attached screenshots of one of the affected comparisons and the IC map used to compute it (which looks good to us). So we are wondering if there is a known reason for this, like these types of IC are always variable between subjects, or if there is a preferred way to detect what is causing these large significant areas in the comparisons.

Any help would be greatly appreciated.