as you are probably analysing modulated data in the GLM, that is local
volumes, it definitely makes sense to correct for total
intracranial volume, or total gray matter volume. Both TIV and total GM
should be calculated from the modulated normalised images, yes,
as only these represent actually *volumes*. (Alternatively the sum of
native probability maps of GM, WM and CSF can be used which should give
exactly the same values).

Thresholding modulated maps: as you write, at first sight, it may see odd
and not logical to threshold the modulated volume maps. However, as they
are based on probability maps that are then corrected by JM for the warping
effects, very low probabilities may represent not so valid estimations,
arguing for a thresholding step to neglect those low prob voxels. As long
as the same rule is applied for all subjects, it is unlikely to make a big
difference for the VBM analysis in the end. For the CSF compartment in
which low prob values and low modulated values sometimes extend far beyond
intracranial area, it can make a big difference (even for SPM5 based CSF
maps), again arguing for thresholding, or alternatively, masking out
extracranial values by a customised ot thresholded wholebrain mask.

To get a feeling for the effect of thresholding it may be worth to try some
thresholds and see what it does to "real" absolute mL of the total
intracranial brain volume and the brain parenchyma fraction for which
values have been published (well, if you have lots of time and want to make
it perfect...)

best regards,

Max Planck Institute of Psychiatry
NMR Research Group
Kraepelinsr. 2-10
80804 Munich
Mail: [log in to unmask]
Phone: 0049-89-30622-413