 |
 |
Hi Matan, Since you want to do FDR, the p-values need to be the uncorrected. I don't think FEAT outputs those, but you can generate them with ztop (e.g., in fslmaths). Then take the maximum p-value between the two contrasts (also using fslmaths). That is, keep the least significant p-value, and apply FDR on the result. I don't think there has been much investigation on conjunctions together FDR, but I believe the above should give an amount of false discoveries that is within the FDR level. All the best, Anderson -- Anderson M. Winkler FMRIB / Analysis Group [ Blog <http://brainder.org/> | Twitter <http://twitter.com/AndersonWinkler> ] On 18 August 2015 at 14:12, Matan Mazor <[log in to unmask]> wrote: > Hi experts, > > I want to detect voxels that are both more responsive to A than B and more > responsive to A than C at the population level (RFX). Seems like the > straight forward FSL way for doing this is to use 'contrast masking' at the > first level (mask A-B with A-C) and then use FLAME for the second level. I > want to use FDR, which makes things more complicated, since it can't > control for the FDR of both contrasts, but only for the contrast being > masked (A-B). > > I couldn't find a way to control for both, so I decided to use (Nichols et > al 2005)'s minimum statistic method (MS/CN) instead. For each subject I > created a new cope image comprised of the minimum value of each voxel > between cope1 (A-B) and cope2(A-C). I couldn't find a reasonable method for > assessing the varcope and the tdof of my conjugated contrast, so instead I > used Simple OLS and for each voxel performed a one sample t test on that > voxel's value. > > I have two questions: > > 1. Does this approach sound reasonable? > 2. Is there a better way? (specifically - is there a way to evaluate the > varcope and tdof of the conjugated contrast?) > > Many thanks, > - Matan >