I just want to make a note about using the average time courses in ROI's. I would advise against computing any sort of correlation matrix between ROI's that used these average time courses. The correlation between an ROI with a large number of voxels and all other ROI's will be artificially inflated due to the smaller standard deviation in the denominator.
I have found that a possible alternative is to bootstrap the mean time course. An equal number of samples is used to compute the variances.
Dan