HMC workflow does not work well on multi-delay data

[tsalo]

Summary

Multi-delay ASL data has different contrasts for different delays, which is causing the MCFLIRT-based HMC workflow to perform poorly.

Two options that we've discussed:

• Replace MCFLIRT with individual ANTs transforms. I started on this in Replace MCFLIRT-based HMC workflow with ANTs-based one #506, but I haven't figured out how to generate RMSD and other outputs that we need.
• NiFreeze

[tsalo]

@effigies has an idea to create references for each of the volume types and registering them to the reference image, and then use those volume-type-specific references as the target for MCFLIRT.

Also, @aptinis will try changing the metric for MCFLIRT to mutual information first.

Another option, from @cookpa and @caugolm, is to create a simple CBF image from the data to use as a reference image first.

[cookpa]

@caugolm said it more precisely than I did, it could be a CBF image or simply a perfusion image - doesn't need normalization to CBF necessarily. Just need to cancel out the spiral ringing.

[tsalo]

Changing the cost function seems to have worked. If anyone reports issues with motion correction, I think we can revisit creating volume type-specific reference images or a CBF image. This is distinct from calculating an early CBF image for coregistration (see #503, #343, and #186).