Mean-signal DKI

[arokem]

The mean-signal DKI model implemented in DIPY (https://docs.dipy.org/stable/examples_built/reconstruction/reconst_msdki.html) has some advantages for computation of kurtosis and related metrics in data with high levels of noise and artifacts. It might be good to enable use of MSDKI derivatives in addition to DKI derivatives.

[36000]

I think we should choose one or the other to avoid too many covariates in our output data. FA/MD/MK already covary enough. maybe we could try to figure out if one is generally more useful? or, if they are useful in different cases, automatically find that and choose a reasonable default?

[arokem]

Oh - I agree. I wouldn't suggest that anyone run both DKI and MSDKI, but I think that it might be good to have MSDKI as an option, especially for cases where the "standard DKI" has a lot of issues (as we have sometimes seen). I don't think this is something that can be easily automated though. Generally, I think that Rafael and colleagues would actually argue that MK from MSDKI should be the default and is inherently superior to MK from the "standard DKI", because it also avoids the impact of fiber configuration. For now, I think that it might be worth adding it as another option (and maybe even warn users who are trying to run both MSDKI and DKI that they will be computing closely related quantities).