At higher fields, there is sometimes quite a lot of intensity nonuniformity,
which can throw off the initial affine registration by quite a lot (because
this does not model intensity non-uniformities). If the initial manual
alignment is very close, then you can sometimes benefit from disabling this
part of the procedure.
Skull-stripping can also help, as the SPM5 segmentation model only has tissue
probability maps for GM, WM and CSF. The model knows nothing about other
tissues in the head, and these can confuse the algorithm.
Best regards,
-John
On Wednesday 17 December 2008 16:48, Philip Kohn wrote:
> Funny, I was having the same problem when I got your email.
> I tried various things, and what worked in my case was to increase the
> number of Gaussians per class for CSF from 2 to 4.
> So that variable will now be 2 2 4 4 instead of 2 2 2 4.
> Maybe the CSF + skull is too variable to be well modeled with only 2
> Gaussians?
>
> Philip Kohn
>
> Ajna wrote:
> > Dear SPMers,
> > I am having difficult time segmenting MPRAGE acquired on 3T with SPM5,
> > and am running out ideas of what might be the problem. Segmentation
> > tends to miss-classify the external cortical gray matter as CSF.
> > Problem is persistent if no manual coregistration of MPRAGE to T1
> > template is not first performed, but even after manual coregistration,
> > segmentation still classifies significant amount of GM as CSF at the
> > brain surfaces. Also, the normalized MPRAGE tends to be imperfectly
> > coregistered to the MNI template, but the fit is only slightly off...
> > My MPRAGE images appear to be high quality, with clear intensity
> > distinction between tissue types. Any advice on how to trouble-shoot
> > this problem would be greatly appreciated.
> >
> > Thank you,
> > -Ajna Borogovac, Columbia University, NY
|