Yeliz - again, depending on how you are implementing this ROI analysis, neither the p- nor t-values should change. Instead, what changes is the threshold. You should certainly interpret this significance cautiously; according to standard statistical practice, if you intended a priori to test this hypothesis in this way (that is, to do an ROI analysis in amygdala), then you are "safe", but if not, then you are dredging. However, there is a logical limit to this reasoning, since in some sense, you could say you were interested in the activity in every voxel in a whole-brain analysis (but few would find that a compelling defense of failing to correct for multiple comparisons). As long as you are clear about what steps you took, there is nothing "wrong" with what you did; however, based on your description, you should certainly describe this analysis as an exploratory one, and the significance level will therefore be understood (at least, by savvy readers) to be more liberal.