Hi Alfonso,

Thank you very much for your detailed response!

I definitely agree with your motivation for switching to a common set of nodes across subjects as QC-FC correlations are very useful/informative (to be honest, it did not occur to me that differently sampled voxels for different subjects would impact these correlations but of course it would).

I think your suggestion to revert to the old method if a non-default mask is defined is a good one. Obviously QC-FC correlations would go back to being confounded from the different voxel sets, but at n=1,000 it should start approaching "truth" right? I had considered another alternative for projects with non-standard templates and/or involving subjects for whom the putative GM reference volume isn't entirely valid (e.g. patients with lesions/tumors/resections): generate a project specific GM reference volume that is simply the intersection of all subjects' thresholded GM probability maps or eroded GM masks and then sample 1,000 voxels from that. On the plus side, this allows the same subset of voxels to be used for all subjects, however 1) perhaps this mask can rapidly get too small if too many subjects with large/varied lesions are included and 2) the QC-FC correlations would then be ignoring FC values in some valid gray matter voxels for most subjects and I don't know that that's a better problem to have than including FC values from non-GM voxels of other subjects...