Hi FSL experts,
I am running a voxel-wise FA skeleton comparison using randomise function. I got the statistic results which I am doubting about.
My project has 3 groups of 563 subjects, which I excluded 80 subjects for incomplete information or site differences. The design matrix has 86 columns, including 3 columns of groups, 3 columns of covariates, and 80 columns of excluded subjects (coded 1 for the subjects to be excluded).
design.mat (7th~86th columns coded for subjects to be excluded)
/NumWaves 86
/NumPoints 563
/Matrix
1 0 0 31 1 3 0 0 0 0 0 ....
0 1 0 25 2 3 0 0 0 0 0 ....
0 0 1 21 1 6 0 0 0 0 0 ....
0 0 0 27 2 6 0 0 1 0 0 ....
....
The design.con file includes 3 contrasts: grp1- grp2, grp1- grp3, grp2 - grp3.
/NumWaves 86
/NumPoints 3
/Matrix
1 0 -1 0 0 0 0 0 0 0 0 ....
0 1 -1 0 0 0 0 0 0 0 0 ....
1 -1 0 0 0 0 0 0 0 0 0 ....
randomise -i all_FA_skeletonised.nii.gz -o TBSS -m mean_FA_skeleton_mask.nii.gz -d design.mat -t design.con -n 1000 --T2
My puzzle is, the max value of tfce_corrp images are so small. The max value in uncorrected p_stats images are 0.992, 0.996, 0.998 for the 3 contrasts, whereas the max corrp value are 0.026, 0.078, 0.606. The range of t_stats are -5.56~4.45, -5.36~4.18, -4.72~4.07. Does these FWE-corrected results looks normal? If yes, what might be the possible reasons for corrp so small? In my impression, the corrp values are not so largely different from the uncorrected p values.
Thank you in advance!
Best,
Xiao
|