Draft MEDIC dynamic distortion correction method

[tsalo]

Closes #36.

This currently runs on some test data, but I haven't evaluated the results.

Changes proposed:

• Add sdcflows.workflows.fit.medic module with MEDIC workflow.
  • I translated vanandrew/warpkit into Nipype workflows.

Still to do:

• Figure out how to make EnforceTemporalConsistency more memory-efficient.
  • warpkit is much faster.
• Compare SDCFlows results to MEDIC implementation in warpkit.
  • The current results look really bad compared to warpkit, but I'm confident I can pin down where they diverge.
• Convert functions used as nodes into Nipype interfaces.
• Add MEDIC option to the field map wrangler.
• Documentation
• Tests

[codecov]

Codecov Report

Attention: Patch coverage is 19.83871% with 497 lines in your changes are missing coverage. Please review.

Project coverage is 66.28%. Comparing base (73d717a) to head (ed0f46c).

Files	Patch %	Lines
sdcflows/utils/misc.py	7.14%	234 Missing ⚠️
sdcflows/interfaces/utils.py	24.51%	117 Missing ⚠️
sdcflows/workflows/fit/medic.py	12.62%	90 Missing ⚠️
sdcflows/interfaces/fmap.py	55.31%	42 Missing ⚠️
sdcflows/utils/phasemanip.py	6.66%	14 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           master     #435       +/-   ##
===========================================
- Coverage   76.43%   66.28%   -10.16%     
===========================================
  Files          32       33        +1     
  Lines        2835     3455      +620     
  Branches      376      416       +40     
===========================================
+ Hits         2167     2290      +123     
- Misses        600     1097      +497     
  Partials       68       68               

Flag	Coverage Δ
unittests	?

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tsalo]

I still need to figure out why the results don't look good, but I was thinking an alternative might be for me to simply integrate warpkit as a dependency and run that directly. Would that be preferable to a translated Nipype workflow?

[mgxd]

If using warpkit significantly improves performance, all the more reason to wrap it rather than reimplement - it'll come with the added benefit of reducing longterm maintenance 😄

One hurdle that would add is a more complex installation - however if/when vanandrew/warpkit#6 is resolved it would be a cinch

[tsalo]

@mgxd I did what you proposed in #438, so I'll close this PR now.