afni.nimh.nih.gov

Certainly the ACF parameters should be estimated from the whole brain residuals -- it is very not a good idea to estimate them from a small region.

Otherwise the normal caveat about small volume correction are still present -- were the ROIs selected via some unbiased reasoning?

Another option to consider, if it applies to your situation, is the new 3dttest++ -Clustsim option, with the appropriate mask.