Hi FSL Experts,
I have created multiple fiber tracts in an atlas that represents controls and a disease population, skeletonised them separately with the TBSS tool, and projected multiple types of DTI scalars to these skeletons.
If I use randomise to perform t-tests on each tract skeleton separately to compare DTI scalar values between controls and disease group, I will get uncorrected and corrected p-values maps as outputs, depending on what type of analysis I use with randomise (voxelwise, cluster threshold, or TFCE). I understand that the corrected p-value maps (outputs of randomise, specifically) have already been controlled for multiple comparisons within tract and then doing a Bonferroni correction on these corrected p-value maps will correct for all contrasts done in my study. Forgive me if this is silly, but now I'm trying to understand I would use FDR instead.
For voxelwise and TFCE, if I wanted to control for multiple comparisons within each tract using the FSL's FDR tool, my understanding is I would run FDR on each *_vox_p_tstat* image for each contrast. What are my options for correcting for multiple comparisons, where the multiple comparisons are now the different contrasts (different tracts, different scalars) in my study? Do I need to combine the outputs of FDR by DTI scalar (all tracts with FA) and re-run FDR on the combination image?
For a cluster threshold analysis with randomise, I know that the *clustere_corrp_tstat* has already been corrected for multiple comparisons within tract. Since this is a cluster based analysis, I'm not sure what to do to correct for multiple comparisons across contrasts. The only thing I can think of is manually running p-values of individual clusters through a separate program like R. What do you suggest?
Thank you,
Joy
|