Hi Matthieu,

Please see below:


On 28 September 2015 at 16:51, Matthieu Vanhoutte <[log in to unmask]> wrote:
Dear FSL's experts,

I have one group and three time-points per subject and would like to assess the evolution according time-points of DTI parameters (FA, MD, ...).

I have looked at the FSL mailing and noticed that similar things have been asking as "longitudinal TBSS", but all these questions were based only on two time-points per subject.

1) Could you indicate to me if TBSS analysis is recommended for longitudinal study of DTI parameters ?

It isn't something that could be formulated in terms of being "recommendable" or not. If you'll do a voxel-based analysis on white matter, TBSS is a recommendation, but this isn't related to being longitudinal or not.

In any case, yes, TBSS can be used with longitudinal studies.

 
2) And if this is the case, which steps of TBSS tutorial should I follow and which steps should I personalize in my particular case of three time-points ?

If you have complete data, simply treat each timepoint as if it were an individual, so that the study-specific template and skeleton is unbiased towards any particular timepoint.

 
3) Concerning the statistical analysis, how should I proceed in terms of inputs and models of the longitudinal three time-points ?

Please, see a mockup of the model here, for 5 subjects: https://dl.dropboxusercontent.com/u/2785709/outbox/mailinglist/design_matthiew.ods

You need to supply a file with exchangeability blocks (EBs) to randomise, with the option -e, so that permutations will happen within subject only. This assumes compound symmetry, i.e., the covariances is the same across all timepoints.

If compound symmetry cannot be assumed, there's another possible trick: assemble design and contrasts for one paired t-test, and prepare the 4D file comparing timepoints 1 and 2 (i.e., T1 and T2) to run with this paired t-test. Prepare another file with another paired t-test for T1 and T3 (the design remains the same, and with the subjects in the same order) and another for T2 and T3 (also same design files). Then run PALM using each 4D file as one input (-i), with one design (-d) and one contrast file (-t), with the EBs file using one block per subject (-eb) and the option -corrmod, to correct for the 3 paired t-tests.

All the best,

Anderson

 

Many thanks in advance for helping !!

Best regards,
Matthieu