when we re-ran a full TBSS pipeline (using the tbss_2_reg -n option) after upgrading from version 4.1.8. to 4.1.9, we got some "macroscopic" differences in the results (visible both in the registered FA images and in the stats results).
We traced the differences back to the tbss_2_reg and tbss_3_postreg stages:
- the average warping computed for each subject is noticeably different between the two versions. This leads to selecting a different "best" target, and hence to different registration results as well as different stats results.
- even when running tbss_3_postreg 4.1.8 on data registered with tbss_2_reg 4.1.9, the computed warp magnitudes are slightly different than what we get in the original 4.1.8 analysis (so even the original registration results differ between versions).
Has there bin any significant change in FLIRT or FNIRT functionality (or any other stages of the anlysis) which may explain the above?
the differences you see are likely to be caused by a minor change on tbss_1_preproc, where the mask that is being used in subsequent flirt and fnirt processing is being made slightly larger (more inclusive) than in the previous version. If you comment out the line reading
$FSLDIR/bin/fslmaths FA/${f}_FA_mask -dilD -dilD -sub 1 -abs -add FA/${f}_FA_mask FA/${f}_FA_mask -odt char
I think you will get identical results to before.
Having said that, I would also like to say this. While I can understand it is annoying when your favourite blob disappears/moves when you upgrade FSL we do strive to continuously improve FSL. This might not always be a monotonous process, but it is definitely the long term trend and one should not tailor ones FSL version to ones data.
Good luck Jesper