Hi Markus,
> as a dummy user i do not understand, why spm is not able to simply mask out
> all blobs outside the brain during the segmentation process while
> normalizing to the template?
Well, I guess it's just a philosophical thing: SPM tends not to assume
things. E.g. unless you tell it that you want some kind of masking,
you get all (non-constant) voxels included.
The segmentation model (a mixture of Normal/Gaussian distributions)
assigns at least small probabilities for all tissue classes nearly
everywhere (i.e. hardly any voxels have absolutely zero chance of
being a particular tissue). The Gaussian smoothing also means voxels
quite far from tissue end up with small probabilities.
If you set an absolute threshold of 0.05 or above, you'll almost
certainly find that stuff outside the brain will disappear. If you'd
rather use a mask based on normalisation to MNI space, then you are
free to try this - just specify an image as an explicit mask. You
might want to try a smoothed and thresholded version of the GM prior
or tpm, or a sum of the GM, WM, and CSF priors. Note though that the
normalisation is not perfect, so such an MNI-based mask might not
match your subjects that well (especially if they have some form of
abnormality or pathology).
I hope that helps,
Ged
|