Dear Xiyao,
I suggest the following: For your SVC use an initial voxel threshold similar to the one of your whole-brain analysis (assuming you report anything related). Otherwise the findings are not due to the >>small volume<< correction only, but also due to a different initial voxel threshold. Commonly accepted voxel thresholds are...
1) FWE-corrected .05 (it seems in this case all surviving voxels are going to be significant on FWE-corrected cluster level p < .05 as well, at least my common observation with human whole-brain fMRI data, dunno about the theory behind - in this case you don't need any SVC of course)
2) uncorrected .001 (combined with a corrected cluster p, say .05 FWE)
Then define your small volume and report the corresponding corrected p-values on cluster level for the SVC. If your corrected cluster p is > .05 FWE, then your SVC didn't show any significant clusters.
If you still want to report this cluster, you can try, but then there's no need for a SVC, as it is NOT significant on cluster level, neither on whole-brain nor on SVC. Instead it is similar to reporting clusters based on an uncorrected voxel threshold and an arbitrary extent threshold of say k > 10 voxels on whole-brain. See for example http://scan.oxfordjournals.org/content/4/4/423 , based on some simulations and a specific dataset they discuss an uncorrected initial voxel threshold of .005 and an extent threshold of k > 10, but note that these authors claim "We are not trying to suggest that P < 0.005 with a 10 voxel extent should be reified as a ‘gold standard’ criterion." Many reviewers might not like an arbitrary extent threshold.
Hope this helps,
Helmut
|