Pvalues adjustment in SnPM t-tests / ANOVA

[saphir746]

Hello developers,

I have been following your work with much interest and been using the SPM / SnPM packages for my own analysis.
I have a range of non-human MRI scans (one per animal) across various groups that I want to analyse for structural differences.

I am unsure what is the best way to adjust for false positives / false discovery rates. I can see that the SnPM package is equipped with FWER / FDR correction functions that can be applied to the p-values extracted from the permutation analyses. However, one of your associated papers suggests that the permuation protocol implicitly adjusts for this:

"Formally, we require a test procedure maintaining strong control over image-wise Type I error, giving adjusted P-values, P-values corrected for multiple comparisons [...] In contrast to [... ]parametric and simulation based methods, a nonparametric resampling based approach provides an intuitive and easily implemented solution (Westfall and Young, 1993). The key realization is that the reasoning presented above for permutation tests at a single voxel rely on relabeling entire images, so the arguments can be ex- tended to image level inference by considering an appropriate maximal statistic."
(Nichols, T. E., & Holmes, A. P. (2002). Nonparametric permutation tests for functional neuroimaging: A primer with examples. Human Brain Mapping, 15(1), p. 6)

What type of permutation is implemented in SnPM (voxel-level only or image-wide)? Based on this, do p-values need to be adjusted to reflect true positives? What about cluster-wise inference?

Thank you very much in advance, I very much look forward to your reply.

Cheers