Hi Elias,

On 5 June 2014 11:26, Elias <[log in to unmask]> wrote:

Dear all,

how would one go about doing SVC in FSL? Am I right assume by creating a binarised mask of the ROI(s) and using it it in randomise instead of a whole brain mask?

what I have done so far is create 12mm ROIs for inferior temporal gyrus, bilateral prefronal cortex etc, with peak coordinates taken from a paper doing a similar analysis as ours. Then run randomise:

randomise -i 4D -o vrPT -d design.mat -t design.con -n 10000 -m prefrontal.nii.gz -x

Yep, that's it. All tests and the correction for multiple testing will happen within the mask only (the term SVC isn't much used in this context, although strictly it isn't incorrect -- after all, it's a correction for small volumes; still, I'd say simply "correction for multiple tests within the mask" or something similar).

Also in doing SVC, which correction method you find is most appropriate? I assume that I should report the corrected p-values (_vox_corrp_tstat) than the uncorrected values (_vox_p_tstat)

Corrected p-values, always.

Finally, are there any assumptions that would make for example the -x (voxelwise) correction more appropriate that tfce -T?

TFCE (-T) tends to be more powerful, but this benefit may be less pronounced with tight masks. The -x can just as well be used (you can include both options in the same randomise run).

All the best,

Anderson

Many thanks,

Elias