This describes the creation of the subcort.mask.1mm.mgz

This volume is not conformed in that it is not 256^3 (which make it
even huger than it is now). This volume matches the output geometry as
when mri_vol2vol is run with --talres 1 and so is compatible with
FSFAST. 

These commands are similar to the ones used for creating the 
mri.2mm/subcort.mask.mgz.

The mask is created based on mri/subcort.prob.mgz which was created
with make_average_subcort. subcort.prob.mgz is the raw probability of
a voxel being in a subcortical gray matter structure based on the
Buckner40. Inevitably, this mask will include more than just
subcortial gray matter structures, but we would rather have the mask
be too big than too small.

# Resample into 1mm space
mri_vol2vol --mov subcort.prob.mgz \
  --s fsaverage --tal --talres 1 --o subcort.prob.1mm.mgz \
  --no-save-reg

Threshold at .05 (meaning that at least 5% of the subjects at 
a voxel must have a subcortical label). Also dilate by 3 to
expand the mask.

mri_binarize --i subcort.prob.1mm.mgz --min .05 \
  --dilate 2 --o subcort.mask.tmp.mgz

Erode by 2. The net result of dilating by 3 then eroding
by 2 is that holes are filled in and the edges are 
a little smoother.

mri_binarize --i subcort.mask.tmp.mgz --min .5 \
  --erode 1 --o subcort.mask.1mm.mgz

rm subcort.mask.tmp.mgz subcort.prob.1mm.mgz

Check against 2mm version of mask
tkregister2 --targ subcort.mask.1mm.mgz \
  --mov ../mri.2mm/subcort.mask.mgz \
  --regheader --reg junk.dat