Dear Natalia,
I'd say there is no consensus for that form of multiple testing when it comes to fMRI. Many papers conduct a bunch of tests at a certain threshold without any adjustments (e.g. ROI analyses on several ROIs, each of them thresholded at .05). Any form of correction is better of course:
1) You could use an FWE correction for the different ROIs / network sets, so instead of an initial voxel threshold e.g. .001 uncorrected you would go with .001/n with n corresponding to the number of comparisons.
2) You could also adjust the initial voxel thresholds based on false discovery rate. See http://en.wikipedia.org/wiki/False_discovery_rate for different procedures. This should be more liberal than 1)
Note that this form of correction has nothing to do with the FWE/FDR procedures in SPM except the mathematical principle, as the SPM procedures control for the number of voxels or resels, but not for the number of conducted tests overall.
Best,
Helmut
|