ENH: Deep source modeling

Describe the new feature or enhancement

looking at roadmap and #6784:

now (since 7.x FS) segment_subregions gives proper subsegmentations (hippo, amygdala, thalamic nuclei, and brainstem), also FS 8.x now runs SynthMorph as part of recon-all so every subject already have subject-to-template deformation field saved. So for morphing problem, instead of trying to inflate surfaces to a sphere, we could just take the saved deformation field and apply it directly to the vertex coordinates and then we can find the corresponding template mesh with nearest neighbour or ICP. no?

I mean these files:

mri/transforms/synthmorph.1.0mm.1.0mm/warp.to.mni152.1.0mm.1.0mm.nii.gz        # subject → MNI152
mri/transforms/synthmorph.1.0mm.1.0mm/warp.to.mni152.1.0mm.1.0mm.inv.nii.gz    # MNI152 → subject

Describe your proposed implementation

if yes, then:

1. we can modify setup_source_space or new setup_subcortical_source_space to create SourceSpaces
2. morph with saved volumetric warp, not spherical inflation
3. and a plotting support in Brain

Describe possible alternatives

.

Additional context

Happy to open PRs for that, but better to discuss first. @larsoner

Start by reading the roadmap section on deep source modeling and issue #6784, then inspect setup_source_space, the proposed setup_subcortical_source_space entry point, and Brain plotting support. Review the listed SynthMorph warp and inverse-warp files. Done would mean agreeing on the design and supporting SourceSpaces creation, volumetric morphing, and plotting for deep source models.