> I could not base the analysis on a previous study or atlas
The regions don't have to be all directly related to your current study. For example, you could define 100 potential regions based on an atlas, and it would be fine to have only a small proportion of those 100 regions that might be related to the current data.
Have you performed group analysis with 3dANOVA2 -type 3 or even pairwise comparisons between any two seeds with 3dttest++?
Gang