ENH: improved brain extraction parameters

[eilidhmacnicol]

Needs testing on a more diverse dataset, but this works very well on 24 fairly isotropic T2w images.

Could test with the original dataset on Sherlock and then upstream the N4 changes to niworkflows as part of nipreps/niworkflows#611 if further improvements aren't needed.

[pep8speaks]

Hello @eilidhmacnicol! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-02-23 09:16:57 UTC

[eilidhmacnicol]

Now running on CircleCI, but ran into an issue that seems to have been discussed already (nipreps/niworkflows#619), so have updated to niworkflows 1.4.0rc4 (see: nipreps/niworkflows#599), but an alternative fix is also on the horizon (nipreps/niworkflows#619)

[eilidhmacnicol]

Update: the pin of niworkflows 1.4.0rc4 didn't help, as the scale = argument needs to become scale_x(as in nipreps/niworkflows#619)

[eilidhmacnicol]

Pinned svgutils to 0.3.1 as recommended as the quick fix in (nipreps/niworkflows#595) while waiting for the more permanent fix to be included in the release candidate.

Now tests are passing (and results look quite good if I say so myself).

@oesteban WDYT? What would be best to upstream to niworkflows? Should I merge first and then refactor?