Hi Matthew and SPMers,
At 07:42 PM 2/21/2005 -0800, Matthew Brett wrote:
> > A number of groups have been applying SVC to VBM in their
> > published studies. I understand that, at least in SPM99,
> > non-stationariness of smoothing could make the SVC approach invalid
> > (see message below).
>
>Very good point. Actually, could I add to this question - it was my
>understanding that the extent correction statistic should not be used
>for VBM because of the non-stationarity problem. I'd be very
>grateful for any comments...
Use of cluster-extent test has been discouraged in VBM data analyses since
the early days of VBM because of non-startionarity associated with warping
during the spatial normalization process. To get around this problem, you
have to measure the image smoothness locally at each voxel, then you have
to adjust cluster extent based on this local smoothness measure. SPM
produces an image called RPV.img (RESEL per voxel image) during an
analysis, which can be considered as the measure of local "roughness" and
can be used to adjust the cluster extent. The problem with RPV is that it
has a lot of variability in it, since it's calculated at each voxel and not
pooled over all the voxels. When you do a statistical inference, this
variability in RPV needs to be accounted for, and reduces the sensitivity
of the test quite a bit.
If non-stationarity is an issue, then I think Keith Worsley's fmristat
package should be able to do this "non-stationarity correction". We tried
to implement the non-stationary correction using a permutation test, but a
typical VBM data set is just too big for our program.
If you are interested in more details, please see
Hayasaka et al., Non-Stationary Cluster Size Inference with Random Field
and Permutation Methods.
NeuroImage 22: 676-687 (2004).
-Satoru
Satoru Hayasaka ==============================================
Post-Doctoral Fellow, MR Unit, UCSF / VA Medical Center
Email: shayasak_at_itsa_dot_ucsf_dot_edu Phone:(415) 221-4810 x4237
Homepage: http://www.sph.umich.edu/~hayasaka
==============================================================
|