John:
We've noticed this as well regarding the gray matter and I'm not sure that
larger ventricles is an adequate explanation. First of all other than
caudate there should be very little gray matter near the lateral
ventricles. However, when comparing normals and patients with various
degenerative illnesses we see huge differences in the ventricular regions
(particularly lateral ventricles) and the changes are not just limited to
the caudate regions. The reason for this anomalous? segmentation is not
entirely clear. Consider that it is usually the case that the ventricular
areas are showing more "gray matter" for normals than for patients with
disease. This means the segmentation algorithm is assigning as gray matter
these periventricular regions in the normal group. Why should this be?
Possible solutions include masking out the ventricles, but the fact that
they show up does make other true areas of difference suspect. What are
your thoughts?
Thanks,
Darren
At 05:14 PM 4/2/01 +0100, you wrote:
>| I'm currently running a VBM analysis of AD and Controls using brain
>| extraction, modulation etc. When I come to the basic models stage I
>| select the following
>| no grand mean scaling
>| threshold masking NONE
>| implicit mask (ignore zero's) YES
>| explicity mask images NO
>| Global calculation OMIT
>|
>| When I look at the mean images of AD's and Controls the mask on the AD's
>| appears larger than that on the Controls (see attached- left images AD,
>| right images Controls).
>| Should I be using a mask in VBM? and If I should can anyone explain my
>| the mask appears different on the two groups when there is only one
>| mask.img.
>
>You should use masking with VBM, but I can't explain why the mask appears
>larger on one group than the other.
>
>Incidentally, I would expect to see apparently more grey matter around
>the corpus callosum of the alzheimers group. This reflects a real difference
>in ventricular size, rather than grey matter. Because of the high contrast
>between brain tissue and CSF, the spatial normalisation works very hard
>at decreasing the sizes of extra large ventricles. It is unable to do this
>without shrinking all the tissue in that region, which has the effect of
>pulling in grey matter towards the centre of the brain. We have started
>to get around this problem by spatially normalising based only on the
>grey matter in the images (by segmenting and matching to the grey matter
>probability images).
>
>
>|
>| Also can anyone explain why when you look at the mask images you get
>| grey areas at the boundaries as well as just black (0) and white (1).
>
>This is because the Display utility shows the images with trilinear
>interpolation by default. The voxels at the edges are grey because
>they are a weighted average of zeros and ones. Try switching to
>nearest neighbour.
>
>Best regards,
>-John
-------------------------------------------------------------------------
Darren R. Gitelman, M.D.
Cognitive Neurology and Alzheimer¹s Disease Center
E-mail: [log in to unmask]
WWW: http://www.brain.northwestern.edus
Voice: (312) 908-9023
Fax: (312) 908-8789
Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
-------------------------------------------------------------------------
|