Hi FSL team,

After running 'tbss_2_reg -n' and 'tbss_3_postreg -S' on a large sample of toddlers, the resulting mean_FA_skeleton and every volume of all_FA had a large dropout in right superior frontal lobe.  I traced this artifact back to a single FA map that had this same dropout.

Here are my specific questions, and I have also detailed my processing pipeline.  I thank you very much in advance!
1) Is it possible that 1 faulty subject with artifact could apply this artifact to all other subjects during tbss_2_reg and tbss_3_postreg?
2) Is it possible to remove a subject (and all associated files) from a TBSS analysis after tbss_2_reg has already ran?
3) Could the troubleshooting attempt in #2 affect the statistical analysis and power to detect direct between-group differences?

Here are the the steps of my processing pipeline.  My questions (from above) pertain to the steps marked with ***:
-I am running all TBSS steps on FSL version 4.1.4., using a quadcore MacPro running OX 10.5.8
-The sample is a between-group design of toddlers N=77, clinical vs. controls. 
-I originally ran tbss_2_reg -n (flag), and it completed after 17 days without crashing.
-I ran tbss_3_postreg -S (flag), and then I checked the alignment of all_FA under mean_FA_skeleton.
-I discovered that every volume (0-76) of all_FA had a large signal dropout in the right superior frontal lobe.  The mean_FA_skeleton also had this artifact.
-After some investigation, I found that the FA map for 1 single clinical subject had this artifact.  I assumed that this one subject was causing all the volumes of all_FA and the mean_FA_skeleton to have this artifact.***
-Rather than removing the 1 subject from the start and re-running tbss_2_reg for another 17 days, I tried the following: I removed all files written during tbss_3_postreg and then removed all files that were written during tbss_2_reg that contained the file name of the faulty subject (i.e. all the nonlinear warps created from the faulty subject during tbss_2_reg).***
-I then ran tbss_3_postreg again, and now every volume of the new all_FA (0-75) did not have the artifact, and neither did mean_FA_skeleton.
-Thus, I proceeded with tbss_4_prestats 0.2 (threshold) and ran randomise with the --T2 (TFCE flag).
-The corrected (corrp) statmap did not show any significant group differences, even at very low (1-p=.60) threshold.***

Thank you so much for your help.

Mark Shen
M.I.N.D. Institute
University of California, Davis