Dear FSL experts,
I am attempting to run a TBSS analysis with the end-goal of running an independent t-test (n1 = 28 and n2 = 20) to find between-group differences. However, after the final stage, randomise gives me very strange results. The first contrast (n1 > n2) shows widespread FWE-significant differences across almost the entire skeleton (albeit no p:s under 0.02). Nothing remotely close to this widespread increase in FA has been seen in any previous study in the same area. Hence, I do not trust my results, especially since the other contrast (n2 > n1) shows nothing. Nada. (Not a single voxel with a p-value below 1).
I have never before come accross this pattern of results before, even when doing preliminary analyses on a subsample of the larger sample I use now. I have followed the TBSS manual and checked each step--at least to the best of my ability. To see if the issue was related to the direct use of the skeleton, I ran randomise on all_FA.nii.gz but the pattern of results came out the same (too many low p-values in the first contrast and too few in the other). I have also tried using both my own skeleton and the FMRIB58-derived one--no difference.
My advisor thinks she has seen this issue before, but cannot recall the specifics. Could it perhaps be that one group is systematically misaligned somehow? Has anyone else gotten the same kind of results? Any tips or speculations would be much appreciated.
Kind regards from Sweden,
Philip Lindner
PhD student, Karolinska Institutet
|