> I have questions concerning the segmentation process.
>
> How are segmented (manually by an expert , with a specific computerized
> method ...) the original images for obtaining binary images basis for
> the probability images (gray.img, csf.img and white.img) ?
I believe that they were segmented using some form of tissue classification
routine. If I was to guess, I would say that a number of voxels were selected
from three different tissue types, the means and variances computed, and
finally belonging probabilities assigned to each voxel belonging to each
class based on the computed means and variances, and the voxel intensities.
Then the brains in the images were masked, so that voxels outside the brain
were set to zero. I think the same mask was used for all images. I made a
slight mistake in the scaling of the probability images when converting them
from their original format to the Analyze format. This is fixed in the next
SPM release.
> What is the meaning and the utility of symetric_gray.img,
> symetric_white.img and symetric_csf.img ?
These images are not used by SPM, but there were a few in-house routines at
the FIL that used them. I didn't mean them to be released with SPM96.
> Another question, I understand that the values of probability images are
> stored between 0 and 1 but why is it the same for the T1.img and T2.img
> templates (and not between 0 and 255)?
I had no particular reason to do this, but I also had no particular reason
not to do it. In the normalization, the templates are scaled in order to
match the images. Occasionally, it is necessary to change the scalefactors
in the template images because of numerical problems when the intensities
in the images are extremely dissimilar, but as a rule it is not necessary.
Regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|