Hi Marina,
I think the CompCor method basically means regressing
out a few principal components from the CSF and maybe
the white matter.
For CSF, consider only using the lateral ventricles.
That avoids including draining veins, and therefore
global signal regression.
See "Example 11" from afni_proc.py help output. It
includes regression of the top 3 principal components
from a FreeSurfer lateral ventricle mask (that _you_
must generate) plus the average eroded white matter
signal. It also uses the eroded white matter mask
for the (fast) ANATICOR step.
- rick