> However, utilizing a .001 uncorrected threshold w/ a extent threshold of 1000 voxels
An uncorrected initial voxel threshold (aka cluster-defining threshold) in combination with a corrected cluster threshold (aka cluster-extent threshold) is absolutely sufficient, you don't have to go with .05 FWE on voxel level. It's just a different way to correct for multiple comparisons, you can do so on voxel level, but you can also do so on cluster level. You should do so >>at some point<< though. In general a cluster extent threshold .05 FWE would typically result in something well below 1000 voxels, meaning some smaller clusters might also reach significance. But of course this depends on your data / smoothness / sample size.
For VBM I would suggest to go with an initial uncorrected threshold of .001 (in the GUI it's the "p value adjustment to control", you would choose "none") plus a cluster extent threshold of .05 FWE and also correct for non-stationarity (different smoothness in different regions of the brain), for that purpose you would have to adjust the settings in spm_defaults.m setting defaults.stats.rft.nonstat to 1. There's no hint inside SPM GUI but the corrected cluster p values on the Results table would be adjusted automatically then.
Best,
Helmut
|