Hi Lynne,
I'm afraid I can't think of any explanations for this, but I'd suggest
looking at the various images with check reg to see if e.g. the
segmentation has been successful, and the smoothed images look okay.
Also check that the mask in the results directory looks reasonable,
and have a look at the ResMS and RPV images there too -- the first is
what the smoothness/roughness is computed from, the second is the
local roughness estimate. (an example of some of mine are attached)
To confirm what you already knew:
Lynne Gauthier wrote:
> I recently ran statistics using SPM with 12 mm smoothed VBM images. When I
> looked at the results, the SPM.mat's .FWHM is a vector of numbers less than
> 1 (I think they should be around 12)?
Yep (e.g. I get 10.7 11.6 10.8 for a 12mm smoothed VBM dataset)
> Also, there are .1 voxels per resel
> (or ~10 resels/voxel). I was under the impression that there should be
> more voxels than resels in the image....
Yeah, I get around 168 voxels/resel.
> so I'm confused by these numbers.
Me too!
> As a result of SPM thinking my data is very "rough", my expected cluster
> size is less than 1 voxel,
Mine's about 16 voxels.
> making any significant voxel show up as
> significant in the cluster-size test. Does anyone have an idea of where I
> might have gone wrong?
Can't think of anything obvious.
> Is my understanding of the numbers I should have
> obtained correct?
Yep, or at least it matches mine...
> Thanks in advance for any advice.
Well, since stationary smoothness-based corrections (cluster-size and
FWE ignore the local RPV estimate) are possibly a bit dubious for VBM
anyway, you might prefer FDR or one of the non-parametric
alternatives. See e.g.
http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0612&L=SPM&P=R44839
Best,
Ged.
|