ENH: Base implementation for phase1/2 fieldmaps

[oesteban]

Putting together the lessons learned in #30, leveraging #52 and #53 (unfolded from #30 too), and utilizing #50 and #51, this workflow adds the phase difference map calculation, considering it one use-case of the general phase-difference fieldmap workflow.

On top of this PR, we can continue the discussions held in #30. Probably, we will want to address #23 the first - the magnitude segmentation is sometimes really bad (e.g. see the phase1/2 unit test).

Another discussion arisen in #30 is the spatial smoothing of the fieldmap (#22).

Finally, the plan is to revise this implementation and determine whether the subtraction should happen before or after PRELUDE, and whether the arctan2 route is more interesting (#59).

[pep8speaks]

Hello @oesteban, Thank you for updating!

Cheers! There are no style issues detected in this Pull Request. 🍻 To test for issues locally, pip install flake8 and then run flake8 sdcflows.

Comment last updated at 2019-11-25 18:15:45 UTC

[oesteban]

Please have a look @Aksoo 👍

[oesteban]

I think the main differences between the phasediff and the phase1/2 now are derived from the quite different mask of the magnitude image:

• Phase-differences
• Phase1/2

Any voices against merging this @Aksoo, @mattcieslak ?

[mattcieslak]

Do you have the actual fieldmaps that we can look at? I'm only seeing the figures on circleci

[oesteban]

Okay, we should now have access to the corresponding NIfTI files resulting from the tests on CircleCI: https://app.circleci.com/jobs/github/oesteban/sdcflows/303/artifacts

They will be available as soon as the tests finish. I will give you a heads-up since I'm very interested in getting this merged ASAP (we are holding a release of fMRIPrep for this and some other changes).

[oesteban]

@mattcieslak - results are up (https://app.circleci.com/jobs/github/oesteban/sdcflows/303/artifacts). LMKWYT

[oesteban]

Lats call before I merge - otherwise, I'm happy to follow up conversation on #22, #23 and #59.

[mattcieslak]

These look very good. The actual values line up closely in the phasediff and subtracted phases

[mattcieslak]

This follows the FUGUE guide, with unwrapping the phase images, then subtracting them.

[oesteban]

Thanks for the review! Very happy to improve this base implementation with your ideas and @Aksoo's. Let's roll this out and build from there.

[kimsin98]

Late response (punted and forgot about this when GitHub broke yesterday) but everything looks good as far as following the old protocol. The refactors greatly improved code organization as well.

I will follow up with my suggestion in #59.