Coregister T2w images

[effigies]

#439 raises the issue of cross-session scaling differences between T1w images, which introduces the possibility that T2w images may also need to be scaled, and not merely realigned to the T1w images.

Reading the recon-all log for uh2/sub-s483, the T2 image is aligned using bbregister, which is 6-dof by default.

It seems reasonable to do a 9-dof registration of T2w images, which would then be saved to derivatives/fmriprep/<subject>/anat/<subject>_T2w_preproc.nii.gz (or should it be <subject>_T2w_space-T1w_preproc.nii.gz?). This re-aligned T2 would also be passed to FreeSurfer.

Questions:

Does FLIRT work well with T2w->T1w coregistrations? bbregister has a --t2 flag to indicate the expected contrast in the moving image; it's not clear from the docs how that should be done with FLIRT. Or should we go ahead and do bbregister --9? Which would mean we need to wait until after the first white surfaces are created (autorecon2) to load the T2.

I'm inclined toward using FLIRT to align all of the structurals at the very beginning, because we run the risk of getting too complicated with the recon-all pipeline.

[chrisgorgo]

Yes - let's try FLIRT or ANTs for this first.

[oesteban]

For inter-modality registration I'd try ANTs first. I can take this one if you want 👍

[effigies]

If you like. It's a little moot at the moment, simply because we were wrong that we needed non-rigid transformations between structurals. But sure, if you want to do this.