Hi, 2086-
The acquired MRI data comes from voxels, essentially setting the minimum, nonzero thickness one *should* be able to measure in a dset, so I see no problem in making a volumetric GM ROI from your surface ROI + pial surface dset + white matter surface dset.
In terms of comparing GM volume between two groups like that.... well... that's a bit trickier, perhaps. Indeed, using the standard SUMA meshes should help this effort---you should be able to generate the "same" (or at least, "analogous") ROI in each subject's native brain space, yes. The larger difficulty I see is: how do you want to compare the ROI volume, once you do calculate it? Do you scale each ROI volume by the subject's total GM volume volume? or by the subject's total brain volume, perhaps? Or by nothing at all? That decision will probably be determined by your research question of interest.
I guess your GM ROI of interest is not a FreeSurfer-estimable one, then? Because then some of that work would be done for you already.
--pt