Dear Venkat,
it is a good idea to correct for non-stationarity in VBM data because the smoothness varies between different brain structures.
In SPM, open spm_defaults.m and simply change defaults.stats.rft.nonstat = 0; to defaults.stats.rft.nonstat = 1; When checking the data with the "Results" button, the reported p-values on cluster-level will be corrected for non-stationarity (takes much longer than "standard" calculation). You don't have to redo the analyses, just switch back to *.nonstat = 0 when you want to look at uncorrected data again.
Note that this might mean that a previously significant cluster of e.g. k = 200 voxels is not significant any more, and a smaller and previously non-significant cluster of e.g. 100 voxels now turns significant (which might be a little annoying if you want to export thresholded t-maps to e.g. Caret for visualization purposes; in this case you will have to apply a threshold of 99 voxels to ensure the 100 voxel cluster is included, and you have to mask the larger but non-significant 200 voxel cluster with MRIcroN manually).
Best,
Gabor
|