Hello,
looking at the attached image I do not think it is not (only) a problem of
default versus study priors.
You seem to segmentate a normalized image ("w"), so theoretically the priors
(default or customised) should be positioned well enough to not produce such
a poor result. However you report the original voxel size - so I guess the
normalised images should have a more standard resolution.
It may be worth checking the following points:
- priors produced correctly and in the same space as the default priors (if in
doubt, better reslice the average of the 8 mm smoothed images with the default
priors as target).
- Also, the value range of the priors should be checked and compared with the
default priors - depends on if you produced validated scripts etc.
- Selected 'yes' in the 'Images normalized?'-prompt? Okay, this is trivial.
- Good/aceptable normalisation result in the four critical cases? Maybe
checkreg all your normalized images?
Bias correction may play a role but does not explain why spm_segmentate
seems to completely misclassify CSF.
Best regards,
Philipp
|