I have a few questions regarding a longitudinal study using TBSS. I have a group of patients who have been scanned at 3 time points pre/pr - before intervention and at 2 times points after intervention (post1/po1 and post2/po2 after intervention) - the time from intervention to po1 and po2 scans varies from subject to subject.
1) With regard to data analysis I have registered all images to the first time point and then proceeded with registration to a target template (using combined warps to ensure images are only written out once). Is this ok ?
2) With regard to model design I have opted for the "tripled t-test" to compare diffusion parameters on a pair wise basis between time points. Is this ok ? The model and contrasts are similar to those on the FEAT web page. That is (example shown for 2 subjects only):
Group Ev 1 Ev 2 Ev 3 Ev 4
Subj1-pr 1 1 1 1 0
Subj1-po1 1 1 1 0 1
Subj1-po2 1 -1 0 1 0
Subj2-pr 1 -1 0 0 1
Subj2-po1 1 0 -1 1 0
Subj1-po2 1 0 -1 0 1
eg Contrast pr-po2 = 1 2 0 0
My questions are:
a) Does one have to correct in some way if one decides to use all 6 possible pairwise contrasts between the 3 time points ? Or does the model take care of that.
b) If one wants to then see how the magnitude of difference in a diffusion contrast (g FA) between time points (eg the contrast shown above) correlates with other co-variates how would one do this ? For example using the contrast above if one also wanted to see if there was a correlation between the difference between pr and po2 scans and the time from intervention to po2 scan how would this best be done ?
3) I have seen some posters from HBM using the mode of anisotropy to distinguish between where changes in FA are due to changes in voxels containing 2 fibre bundles to 1 fibre bundle (causing an increase in the mode to more positive values) - is this interpretation only valid in areas where there are crossing fibres - how does one interpret such changes in the mode in areas where crossing fibres are less likely ?