Hi Sam,
If you are breaking this down to single subject
ROI pair correlations, then indeed my concern about
overlap does not apply.
I even wonder about the need to blur at all. Since
you are presumably averaging over each ROI before
computing correlations, the only thing blurring does
is introduce signal from outside each ROI. Taking
the average is like a big, within ROI blur already.
Why blur, even across gray matter?
When we apply ROI averages as regressors of interest
or no interest, it is done before blurring. I would
expect the same to be appropriate for generation of
correlation matrices.
Just to add something else to ponder...
- rick