adopt msmt csd

[ebrahimebrahim]

Since we want to do whole-brain registration using FOD-derived information, it may make sense to do msmt csd[^1]. With ABCD imaging we have the shells to do it (5 b-values including 0) . This way we'd have something in the CSF and GM regions rather than just WM. We would obtain response functions for each tissue type^2 and then average those over the population^3 and use those in an msmt csd solver^4. The tissue volume fraction map that comes from this may be appropriate to use as a first step of registration, before using the white matter FOD coefficients at increasing spherical harmonic order to iteratively refine the registration.

[^1]: Jeurissen, Ben, et al. "Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data." NeuroImage 103 (2014): 411-426. https://doi.org/10.1016/j.neuroimage.2014.07.061

[ebrahimebrahim]

Keeping things mostly single tissue with the focus on WM, but #5 takes measures to take what advantage we can take of multiple shells. As of merging #5:

• the mrtrix processing does multishell single tissue essentially
• the dipy processing takes advantage of the multiple shells. for response function estimation it only includes shells with b values <= 1200 but not higher b-values because apparently it isn't good for DTI. dipy uses DTI to represent estimated response functions. i essentially followed the advice here.