thanks Steve for the reply- it actually makes sense
I re-run the analyses without including
the motion parameters as confound EVS, then the FEAT results
were very much the same in FSL4 and FSL5, across Ubuntu and Redhat
i concur is interesting to see how tiny differences in motion confounds
can have such effect in Z stats
hence it appears to me that correcting for motion in that manner is quite
an arbitrary thing, and we should avoid if possible......
alternatives are using denoising ICA outputs on individual subjects but this
can be time consuming and prone to error if done by eye based on some heuristics
I may ask then what your opinion is about running group ICA, then identify the motion/artifact
components at the group level and then denoise the individual subject data based on that ICA
group component- guess this seems like a feasible approach?
Cheers
|