> we would like to repeat VBM studies using SPM5 instead of SPM2.
>
> If I understand it correctly the unified segmentation method is implemented
> in SPM5 - this makes things a lot easier. However, what does this mean for
> customising priors?
>
> 1. Should we perform one run with standard priors and a second run with
> averaged, normalised and smoothed seg1, 2, and 3 (=customised priors)?
Personally, I wouldn't bother with custom priors for SPM5 - unless you plan to
do it properly and use hundreds of subjects in order to get a good
representation of the population you plan to study.
You may be able to do some sort of model comparison (to see what priors give a
better segmentation model) by comparing the log-probabilities. These are not
explicitly saved, but a few tweaks to the Matlab code would do it. This
wouldn't be a proper model compariso framework, but it may be able to allow
you to compare different tissue probability maps to see which ones best
explain the data.
>
> 2. Should a wholehead template be produced and exchanged, too?
It isn't needed by the segmentation in SPM5.
>
> 3. In 'optimized' vbm normalisation was driven by grey matter - is this
> concept out of date with unified segmentation process?
The segmentation in SPM5 combines tissue classification with matching GM, WM
and CSF to tissue probability maps. It does this within a proper model -
which wasn't really the case for "optimised VBM". There are still some small
things that I would like to improve though. In particular, the model
contains no information about tissue outside the brain.
Best regards,
-John
|