You could disable the bias correction (or set the regularisation to be very
high), and change the number of Gaussians per tissue class. These would make
the behaviour more like thatof SPM99. You could also smooth the tpm data by
about 8mm, which would also make the behaviour more SPM99-like.
Best regards,
-John
On Tuesday 22 April 2008 13:11, Thomas Moorhead wrote:
> Dear Spm'rs
> Further to my query on SPM5 segmentation and control of smoothness I
> have been comparing segmentation results obtained from SPM5 and SPM99.
> My observation is that the SPM5 implementation is more likely to
> partial volume at any given intensity value. This is not necessarily a
> problem if all scans come from the same or equivalent platforms and
> inhomogeneity gradients are low. However I have found that where
> platforms differ that the distance between segmentations is less if I
> employ SPM99 rather than SPM5 as the tissue classifier. I am wondering
> if there are facilities within SPM5 that would allow me to manage the
> nonlinear characteristics of its classifier.
> Bill Moorhead
> Psychiatry
> Edinburgh
|