I’m trying to track specific fibers in population template space to use as ROI masks for fixel-based statistical analysis.
I was wondering if it would make sense to co-register a T1 MNI template to the population template and then create a 5tt file from the co-registered MNI T1 and use the 5tt file for anatomically-constrained tractography in population template space?
I have come across the use of co-registered atlases for ROI-specific fixel-based analysis, but is using MNI T1 for ACT a step too far?
This question has come up a couple of times in the past, but I don’t yet have a definitive answer. Good opportunity to be the tip of the spear!
There’s quite a broad range of way that one could derive a 5TT image intended for population template tractography. But I’ve never gone to the effort of testing them to see what might or might not work. Ultimately, if you can generate such an image, in a data-driven way, it might be better than not doing such for certain applications: for whole-brain FBA the statistical enhancement is strongest proximal to each fixel so having the far portion of a streamline not quite reaching the cortex doesn’t really matter that much; but for WM bundle segmentation it might have an effect.
If you have EPI distortion correction & inter-modality alignment, I’d maybe try warping the T1 images of your population subjects based on the FOD-registration-derived warps, see if that produces something that could feasibly be segmented. Also bear in mind that 5ttgen fsl is not the only way to produce a 5TT image, it’s just a script that’s provided because It’s what I made for myself at the time; you can do manual segmentations, or run any other segmentation algorithm, and massage it into the 5TT format. Segmentation of some other template may introduce too much dependence on the coregistration of that template with your population-specific one.
One esoteric thing to consider is that the way the priors in ACT are constructed is based on presumption of hard boundaries between tissues (use of image space is just a simple & computationally efficient way of representing such). In template space, if you have variability in gyrification, while the registration will do the best it can to provide an average space & subject alignment to such, the idea of “this voxel is white matter and this adjacent voxel is grey matter” is not so apt. While I prefer this construction for single-subject tracking, for population template tracking, probabilistic priors might make more sense.
Long-term, my hope is that the combination of multi-tissue CSD with multi-contrast registration will allow for generation of tissue partial volume fractions based on the input DWI data that can be utilized for ACT. But I’ve not put data through such a pipeline to see…
Food for thought; keep me up to date if you decide to give it a go!
Rob
