Noticeable difference in scanner-computed fieldmap results since LTS

[effigies]

What happened?

We have a dataset with sbrefs, bold and direct fieldmaps. Correcting with sdcflows 1.3 (fmriprep 20.2.7 LTS) gets sensible-looking results. Correcting with sdcflows 2.6 (fmriprep 23.2.0a3.dev) does not. See screenshots attached.

This is a flag to dig into precisely where our estimates diverge. I have a suspicion that we shouldn't be mode-centering the fieldmap when fitting a spline field, but I'm not convinced that's enough to explain this.

There also seem to be regions suppressed by the Jacobian modulation. This seems like it could be explained by an effective shift in the fieldmap, lying to the resampler about where there is suppressed and increased signal.

What command did you use?

* fMRIPrep 20.2.7: `fmriprep /data /out participant --notrack --skip-bids-validation -vv --mem-mb 48000 --ignore sbref --fs-subjects-dir /opt/subjects -w /scratch`
* fMRIPrep 23.2.0a3.dev: `fmriprep sourcedata/raw . participant --notrack --skip-bids-validation -vv -w /data/scratch/fmriprep-ng-uh2aim2 --mem-mb 48000 --ignore sbref --level minimal --fs-subjects-dir sourcedata/freesurfer`

What version of the software are you running?

multiple

How are you running this software?

Other

Is your data BIDS valid?

Yes

Are you reusing any previously computed results?

FreeSurfer

Please copy and paste any relevant log output.

No response

Additional information / screenshots

LTS

Before	After

Next

Before	After