Hello,
> This happen for any range between 0.05 to 0.001
> with 0.05 - one huge cluster for 0.001 - 3 very very very big clusters (which pass fdr\fwe).
> (there are many more smaller clusters which doesn't pass the cluster level FDR\FWE correction so i neglected t mention them)
For a standard cluster level correction, the cluster-level p values
are not valid with liberal voxelwise p values. This has to do with the
sort of shapes/features that show up when you threshold using p < .05
vs. p < .001, for example. The general rule of thumb (which I believe
has been backed up by some simulations) is that p < .001 or more
stringent is a valid voxelwise threshold when using cluster level
analysis.
For VBM data, you will also need to use a nonstationarity correction,
either the ns toolbox (which may no longer be available) or changing
the SPM defaults. There are other posts on this, but it may change
your results. (These account for the fact that the smoothness of the
images are not the same in every location.)
I think Donald's suggestion that you may have group differences in
intercranial volume or total brain/GM volume are a likely source of
your big differences. For intercranial volume and head size, you might
want to see:
Barnes J, Ridgway GR, Bartlett J, Henley SMD, Lehmann M, Hobbs N,
Clarkson MJ, MacManus DG, Ourselin S, Fox NC (2010) Head size, age and
gender adjustment in MRI studies: a necessary nuisance? NeuroImage
53:1244-1255.
and for total GM volume:
Peelle JE, Cusack R, Henson RNA (2012) Adjusting for global effects in
voxel-based morphometry: Gray matter decline in normal aging.
NeuroImage 60:1503-1516.
It's probably worth noting that just because you are seeing widespread
differences doesn't mean that this is incorrect in some way; it just
depends on the question you are asking. It can, however, make your
results difficult to interpret.
Although it's hard to say for sure, it sounds like a good approach
would be to stick with the p < .001 voxelwise threshold, and then
consider the clusters that pass a nonstationarity-corrected cluster
level threshold. If those clusters span more than one brain region,
you just have to realize that it's hard to be very precise about the
location of your effect, and this ambiguity should be reflected in
your discussion of the results.
Hope this helps!
Best regards,
Jonathan
--
Jonathan Peelle, PhD
Assistant Professor
Department of Otolaryngology
Washington University in St. Louis
Office: (314) 362-9044
http://peellelab.org || http://jonathanpeelle.net
|