The priors for the segmentation model are not fully specified as there are no
tissue probability maps for skull, muscle, fat, skin, and other non-brain
tissue. For this reason, the approach needs to be modified in order to
exclude voxels containing air. Without excluding them, the model would
attempt to stretch the tissue probability maps to fit the whole head, rather
than just the brain.
The original solution was to threshold the image so that low intensity voxels
were excluded. For most sequences, this threshold is between the intensity
of air and that of CSF - but I guess you have very low intensity CSF in your
data, so much of the CSF is also excluded.
More recently (i.e. Tuesday), a new set of updates for SPM5 were released.
These included a modification to the way that voxels far from the brain are
excluded from the segmentation. I would suggest trying this updated version,
which I think (hope) will work slightly better for segmenting CSF.
Best regards,
-John
> I have an anatomic MRI volume (T1-weighted) that I segment with SPM99 and
> SPM5.
>
> On the CSF volume the results are totally different despite the fact that I
> have used the same options and the same MNI priors.
>
> However, for the white and grey matter the results are exactly the same.
>
> The CSF segmentation is not correct enough with SPM5 (empty ventricules for
> example). See attached picture.
>
> I do not know why?
|