Hello,
here is another opinion on your question:
What worked fine is - in principle - adding up mwc1001, mwc1002 and mwc1003,
however, as mentioned, the CSF part is the problemaic one.
For this purpose we produced a kind of customised 'tiv' mask by just adding up
ALL probability maps including the CSF maps (unmodulated) just to get a mean image
of the brain plus CSF shape. We then thresholded this mask at 0.5 (to get a binary mask)
to cut off the dura parts etc.
0.5 was a good guess, we checked this value by overlaying it with number of normalised
whole heads to see if the border is right between outer CSF and bone.
Some adjustment may be necessary.
By multiplying this mask with the GM, WM, and CSF compartment we get corrected
values, which add up to a 'corrected' TIV value.
A couple of plausibility tests (men/women, percentage GM etc) showed that the velues correspond
eith published values.
br,
Philipp
|