Both of these segmentation failures are due to poor initial affine
registration. Essentially, if the initial alignment with the tissue
probability maps is not so good (maybe because of intensity
nonuniformities, large initial misregistrations, or incorrect voxel
sizes etc) then the kinds of results we see here may be expected. The
algorithm behind the Segment button does not include any knowledge of
what would be expected outside the brain, as the tissue probability
maps are only of GM, WM and CSF. One of the effects of this is that
the initial affine registration may be less robust.
If you use SPM8, there is an alternative segmentation routine
available from among the toolboxes (with the unimaginative name of
"New Segment"). This implementation includes some additional
knowledge of non-brain tissues, so is generally more robust. It will
eventually become the default segmentation routine, but only after I
have written some more code to use the deformation fields it produces
for spatial normalisation.
Best regards,
-John
On 23 October 2010 16:10, Courtney Brown <[log in to unmask]> wrote:
> Hello,
>
> I am having some difficulty with the segmentation step within SPM8. I am
> looking at healthy young adults completing fMRI paradigms. After outputting
> grey matter and white matter segmented files from the anatomical image, I
> have gotten strange results (please see attached jpeg images). This does not
> happen each time I run segmentation, but I have gotten similar output files
> on several occasions. I have tried running segmentation with the cropped
> anatomical image as opposed to the original anatomical image. The images
> above involve the SPM8 segmentation default settings. I have also tried to
> adjust some of the segmentation defaults (higher/ lower Bias regularisation,
> higher/ lower Bias FWHM), but have gotten similar results in each case.
>
> My question is: Does anyone have a sense of what would cause these types of
> images being outputted after the segmentation step. And furthermore, what
> would be the best resolution to this issue in order to adequately preprocess
> fMRI images for analysis?
>
> Any input would be greatly appreciated.
>
> Thank you,
>
> Courtney Brown
>
> Graduate Student
> Department of Psychology
> University of Georgia
|