On 12/21/2020 8:45 PM, Kayle Sawyer wrote:
External Email - Use Caution
We are working on a manuscript revision for an fMRI project that was analyzed with FS-FAST 5.3 years ago. We used mri_glmfit, and then we ran mri_surfcluster (instead of mri_glmfit-sim) with a vertex-wise threshold of p<0.001 and a cluster threshold of 100mm^2. We had applied 8 mm smoothing at the group-level, and the voxel size was 3.125x3.125x5 (resampled to fsaverage surface and to MNI305).
1. One reviewer asked us to provide a rough estimate of the family-wise error rate (FWER) that a cluster threshold of 100mm^2 is equivalent to, or alternatively, to re-analyze everything with mri_glmfit-sim permutation. It would be simpler to provide the FWER estimate. To get the FWER estimate, we referred to Woo et al (2014). However, their simulations were for volume analyses, not surface analyses. What is the FWER (roughly) for a primary threshold of p<0.001 and cluster threshold of 100mm^2? (Obviously, we are hoping for a number less than 0.05.)
Why not run mri_glmfit-sim and just see what the p-value is for that cluster?
2. For the volume, we used mri_surfcluster with a primary (voxel-wise) threshold of 0.001 and a size threshold of 300mm^3. If I’m understanding correctly, Figure 2 from Woo et al. (2014) suggests a FWER p<0.05 might roughly correspond to a cluster size threshold of 200 to 985 voxels (1600 to 7880 mm^3), depending on intrinsic smoothness. That would likely put our threshold of 300mm^3 well above p<0.05. Is this correct?
Again, you can just run mri_glmfit-sim. I would use the version 7 for this one though as I think the volume-based GRF might have changed since 5.3. You can do the same for the surface-based above, but I think the 5.3 version should be ok
Thanks,
-Kayle
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer