> I am having no luck in obtaining clean segmentations, in particular much
> of the retrobulbar anatomy is being included with the gray matter, which
> alters my gm/wm/csf proportions. I have followed advice from the archives
> regarding creating my own study-specific apriori maps
> (the 'make_template.m' script), with even worse results (I am using 30
> subjects, which should be enough to generate usable maps).
>
> My mprage scans were obtained on a 3T scanner, and I understand that the
> apriori maps included with SPM are based on scans from a 1.5T scanner.
> Could this difference be responsible for the poor gray segmentation I am
> seeing? Or is this related to the mprage sequence (or both the sequence
> and field strength)? Furthermore, is anyone able to share any advice in
> creating study-specific apriori maps.
The scanner sequence, field strength, image artifacts etc are largely
irrelevant to study specific tissue probability maps, as these should relate
to the spatial distribution of different brain tissues. The use of study
specific tissue probability maps is more relevant to segmenting unusual
populations of subjects.
In SPM2 and SPM99, study specific templates are more important for different
image contrasts etc. The reason for this is that they may allow a more
accurate registration, allowing a better superposition of the tissue
probability maps on to the image to segment. The initial registration is
based on the mean-squared difference between template and image. If the
contrasts differ, then this objective function is suboptimal. If you do have
a study specific template, then the tissue probability maps must be in good
alignment with it.
Personally, I wouldn't use the SPM2 segmentation. I would go for the one in
SPM5b (but note that the output is in NIFTI format, so can't be used with
SPM2). In SPM5b, the image contrast is less important for the matching, as
this is done directly with the tissue probability maps.
Best regards,
-John
|