Hello dear FSL users!
Would anyone be as kind as to share their wisdom on TBSS statistical
analysis with me? I am new to both FSL and TBSS and wanted to check if it's
possible to do what I want to do.
I am interested in testing for a correlation between FA (and mean, axial and
radial diffusivity) and a behavioural measure that I don't think necessarily
is parametric by its nature.
So far I have only been able to test for the correlation using randomise
because I'm not aware of equivalent nonparametric methods. I'm pretty sure
that the design is fine as it is but the actual stats command is what I'd
really appreciate some help with. The command I've used to do the stats so
far is:
randomise -i all_FA_skeletonised -o newfilename -m mean_FA_skeleton_mask -d
design.mat -t design.con -n 5000 -x -V (for voxelwise stats that I then
threshold appropriately using fdr)
OR
randomise -i all_FA_skeletonised -o newfilename -m mean_FA_skeleton_mask -d
design.mat -t design.con -n 5000 --T2 -V (tfce thresholded images)
Q1: How would I modify the above to do nonparametric stats instead?
The other thing that puzzles me is thresholding, and hence the following
question.
Q2: How does the thresholding such as the above relate to thresholding done
in fslview when looking at the result? That is, once I've thresholded the
results using tfce or fdr correction, do I still have to adjust the
threshold in the fslview window e.g. from 0.949 to 1 to get to p< .05 or
0.999 to get to p< .01 ?
Sorry if these questions seem silly, but I would very greatful for any help.
Kind regards,
Kirsi
|