[MINC-users] Classify

[Jon Erik Ween]

Dear Minc-users!

I'm trying to do some automatic segmentations and VBM on brains with white
matter and hyperintensities and have noticed that "Classify" always miss-labels
deep grey structures as white matter. The scans are 3DT1, intensity corrected
with N3, skull stripped with mincbet and registered in MNI space with MRItotal.
Lesions are masked with a manual lesion mask in the same space. I've tried to
cut out excessive neck-tissue and the zero-pad volumes (as in INSECT, though
running INSECT directly on native scans bombs out in every case). After this
process I notice the following interesting things, in addition to the lack of
segmented deep grey:

1) there is still quite a range of total brain volumes, even though all brains
have been "transformed" to normal space

2) there is still quite a range voxel intensities from brain to brain, even
though all brains should have been "normalized"

3) csf is often classified as background (ie: "holes" in ventricles)

4) There is clearly a lot of variability brain to brain due to atrophy and
lesions, though, for the most part, cortical grey seems appropriately labeled
(even though the deep grey is missing). Some scans clearly fail to
normalize/register/classify.

Any hints how to optimize this? In particular: is there some method for
reviewing raw data to look for things that would through the process off?

I'd appreciate any suggestions!

Cheers

Jon