basically, TIV is the sum of total GM, total WM and total CSF (with CSF
containing the internal and external CSF spaces).
You can calculate the 'total' of an individuals GM, WM or CSF from the sum
of its segmentated maps in native space, or the sum of its modulated maps
in normalized space - these results should be equivalent (by definition).
So, simple in general, but one significant problem arising:
The CSF maps a considerable amounts of voxels that are misclassified as CSF
and that lie outside the intrancranial vault (mostly parts of the skull).
We (here, locally) use a 'tiv' mask to simply mask these voxels out of the
CSF maps (and the GM and WM maps, too) and by this make the total CSF value
Such a 'tiv' mask can be generated in the following way (just one solution,
of course): you add up all (modulated or non-modulated, does not make a big
difference here) normalised GM, WM and CSF images to a preliminary TIV
image (of your study sample, as many subjects as possible), and then find a
threshold that gives you a nice margin that just follows the inside of the
skull (e. g. use the rcolin brain for comparison) but includes the external
CSF spaces and the basal cisterns. Alternatively a sume of the segmentation
priors can be used to create such a mask. So, assuming your mask is
labelled e. g. 'tiv_mask.nii' (i4), your tiv image of a subject can be
calculated using the imcalc function:
tiv_image = (i1+i2+i3).*i4
(with i1, i2 and i3 being modulated GM, WM and CSF, respectively, and i4
being the mask).
The total of this tiv_image would be your TIV.
Also, see the link
to scripts for this purpose (although I could not tell you how Christian
Gaser's script handled the problem of the rather unreliable margins of the
Max Planck Institute of Psychiatry
NMR Research Group
Mail: [log in to unmask]