I have DTI data for subjects in 2 groups (placebo and drug) with two measures each. Subjects had a scan before and after treatment (or placebo). I tried following the "ANOVA: 2-groups, 2-levels per subject (2-way Mixed Effect ANOVA)" example on the FSL documentation page (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#ANOVA:_2-groups.2C_2-levels_per_subject_.282-way_Mixed_Effect_ANOVA.29).
Since this model cannot be used in randomise due to the way it needs to be permuted, I followed the instructions for computing the paired differences within-subject via fslmaths and then did a simple two-sample t-test to test whether the run1-run2 difference differed between the two groups in randomise.
My question comes with how to do this with DTI/TBSS data since it does not seem as straight-forward to get within-subject differences on skeleton FA data. In my first approach, I ran TBSS all the way through tbss_4_prestats to create the all_FA_skeleton for each subject and condition. I took each subject's two skeletons (pre-treatment scan and post-treatment scan) and subtracted to get the different. I merged these files again to get a new all_FA_skeleton file and do randomise for group comparisons. Is subtracting skeleton's a good approach? It seems like data would be lost since the skeleton's are a projection of data.
I then also tried to take the within-subject differences after part of tbss_3_postreg had been run. So I subtracted each subjects pre and post registered FA maps (SUBJECT_FA_to_target.nii.gz), and the positive values to look at run1 - run2 and the negative values to look at run2 - run 1. For both of these maps (run1-run2 and run2-run1) for each subject, the values were between 0 and 1. Then I re-calculated the FA_mean and FA_skeleton. I used a really low threshold of 0.05 for the skeleton. I am not sure if the skeletonization will work correctly on these new values. The results did not look very good.
What is the best approach to use randomise to look at the group and condition differences for DTI TBSS data?
|