TBM with antsLongitudinalCorticalThickness.sh

[Nazuna23]

Dear community,

I would like to perform a TBM on the results of the antsLongitudinalCorticalThickness.sh pipeline. I have two acquisition time points for each subject t1 and t2.

I combined the transformation files of each single timepoints for each subject to the subject template and combined these with the transformation of the subject template to the group template. I therefore used ComposeMultiTransform with the .warp-files and -affine.mat-files from each transformation (subject —> subject template and subject template —> group template).

As a next step I calculated the LogJacobians for each subject for this composed warp-output via CreateJacobianDeterminantImage.

Then I subtracted the composed LogJacobians of t2 minus t1 for each subject and smoothed the resulted difference Log Jacobian t2-t1 and would like to perform a permutation analysis.

However, I am not sure if this is the correct procedure or if there is another way to recommend with the longitudinal pipeline.

Does anyone have an advice for this analysis? Help would be greatly appreciated. Thank you very much in advance!

[cookpa]

This is an interesting question.

If you compose the transforms in as you suggest, you will get a composite transform including both affine and deformable volume changes. That's not necessarily wrong but it matters for interpretation of the results, and is different to how many TBM studies are done, using only the deformable volume changes in group template space.

Some pipelines make the intra-subject registration rigid, then there's only one deformable transform (single subject template to group template). The longitudinal changes are still reflected in appearance differences (eg, decreased GM probability).

[Nazuna23]

Hi,

Thanks for the reply! So would you suggest to only use the SubjectToGroupTemplateLogJacobian.nii.gz for the TBM? I now performed a subtraction of these for each time point (Follow-up minus Baseline of the SubjectToGroupTemplateLogJacobian.nii.gz) and then smoothed the result and performed a permutation analysis. However, the results do not make sense and look completely blurry/noisy, so I think I am missing something.

Is there a possibility to use the antslongitudinalcorticalthickness pipeline for a longitudinal TBM? Or do I need to step back and work with the coregistration of the follow-up anatomy on the baseline anatomy and use the resulting transformation?

Thank you very much in advance!

[cookpa]

I would use the provided Jacobians in the single subject template space, because these just reflect intra-subject deformation. If you directly register followup to baseline, the solution will be biased.

For group analysis, the computed jacobian differences can be warped to group space.

[gdevenyi]

Chiming in here to offer my pipeline, https://github.com/CoBrALab/optimized_antsMultivariateTemplateConstruction which offers a longitudinal DBM/TBM using the same principles as ANTs template construction.