FIX: Generation of RAS displacements fields from VSMs

[oesteban]

After much thought, I have come to the conclusion that the VOX2RAS affine must be applied in all cases, regardless of whether the dataset is oblique or not.

After all, the VSM (voxel-shift-map) is no more than a regularly gridded field of voxel coordinates. The transformation between voxels and mm is biyective and univocal, and formalized by the affine matrix of the image.

This PR simplifies the calculation, and theoretically should resolve most of the issues we are experiencing when resampling data.

Resolves: #218.
Resolves: #236.

Related: nipreps/fmriprep#2210.

[oesteban]

@mgxd this (in combination with the latest fMRIPrep integration) should fix things for your dataset - please let me know

[oesteban]

I'm writing some unit tests, just to be sure this is what we need.

[effigies]

This makes sense to me. Made some very small suggestions, but feel free to reject without argument.

[oesteban]

I'm writing a test with a numerical phantom. I have been able to compare the correction with our B0FieldTransform object (i.e., in voxel space) and with antsApplyTransforms using the displacements field our object generates. With RAS, canonical affine it works. Will configure the test to explore obliquity, flips and axis ordering.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 85.28%. Comparing base (6dbf688) to head (d9c142b).

Files with missing lines	Patch %	Lines
sdcflows/transform.py	93.75%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #237      +/-   ##
==========================================
- Coverage   85.31%   85.28%   -0.04%     
==========================================
  Files          25       25              
  Lines        1771     1767       -4     
  Branches      205      205              
==========================================
- Hits         1511     1507       -4     
  Misses        230      230              
  Partials       30       30              

Flag	Coverage Δ
travis	85.28% <93.75%> (-0.04%)	⬇️

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.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[effigies]

Approving again. Very convincing test!

[oesteban]

Very convincing test!

Yup, it doesn't tell you that we have a good implementation of fieldmaps, but we are safe to say that, once you have the fieldmap our tool and ANTs will apply the transform the same. So it indeed resolves the unit test :)

Merging in, we can include @mgxd on his data later down the line.