ENH: Use ANTs DenoiseImage before conforming anatomical images

[effigies]

Closes #22.

[mgxd]

I think overall this is a nice addition.

I don't suspect it will add too much (in nibabies it takes about ~4mins) but have you tried to see how it affects the runtime of any test data?

One conceptual question - will this have any effect on confounds?

[codecov-commenter]

Codecov Report

Patch coverage: 50.00% and project coverage change: -0.03 ⚠️

Comparison is base (b6c4365) 28.49% compared to head (a8c70ae) 28.47%.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #337      +/-   ##
==========================================
- Coverage   28.49%   28.47%   -0.03%     
==========================================
  Files          19       19              
  Lines        1351     1352       +1     
  Branches      196      196              
==========================================
  Hits          385      385              
- Misses        952      953       +1     
  Partials       14       14              

Impacted Files	Coverage Δ
smriprep/workflows/anatomical.py	16.66% <50.00%> (-0.10%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[effigies]

The impact on confounds would be in the definition of white matter/CSF ROIs, if those change. In subjects where the segmentations are kind of salt-and-pepper, we might end up with larger ROIs for those and CompCor.

I did run on some test data and it took ~40sec on 16 cores. So should be <3min on 4 cores.

[diramigo]

So the output of ANTs DenoiseImage is used for recon-all, right? does this mean that using an existing output of recon-all for smriprep could lead to a very different result? is this version of smriprep already in fmriprep?

[effigies]

This is in 23.1.x. fMRIPrep has always done FreeSurfer a little differently. We have passed INU-corrected T1w images and BOLD masks pre-computed by antsBrainExtraction.sh (or, now, a rewrite of that script as a nipype workflow). And for quite some time we've been pre-merging multiple T1w images, rather than leaving that to FreeSurfer.

If you would like to use vanilla FreeSurfer, pre-run it and pass it to fMRIPrep with --fs-subjects-dir.