mri_robust_template: Return code: 251

[sentry]

Sentry Issue: FMRIPREP-8GP

merge: Return code: 251

[effigies]

Conformation failure:

ERROR: MultiRegistration::loadMovables: images have different voxel sizes.
  Currently not supported, maybe first make conform?
  Debug info: size(4) = 3, 3, 3.3   size(0) = 3, 3, 3

[fliem]

Hi @effigies
I am getting the same message with different t1w resolutions between sessions. t1w_merge is stumbling over

ERROR: MultiRegistration::loadMovables: images have different voxel sizes.
  Currently not supported, maybe first make conform?
  Debug info: size(2) = 1, 0.94, 0.94   size(0) = 1, 0.93799, 0.93799

in the Conform interface, the image is not rescaled or resampled because of the atol of 0.01 mm. What's the reasoning behind this tolerance?
Thanks.

[effigies]

The reasoning was that for very small changes in voxel sizes, resampling actually induced distortion, so a minimum difference was needed.

I'm surprised that FreeSurfer considers this a sufficiently large difference. I would say that either we should check their source code and match our tolerances, or you could pre-resample. For a .001 mm change, simply resetting the pixdims without resampling seems safe.

WDYT?

[fliem]

Freesurfer tolerance is 0.001 (mm, I assume)

What do you think about adding this here, i.e.:

        atol_fine = {'meter': 1e-6, 'mm': 0.001, 'micron': 1}[xyz_unit]
        fix_zooms = not np.allclose(zooms, target_zooms, atol=atol_fine)
        if rescale or resize:
            ...
        elif fix_zooms:
            reoriented.header.set_zooms(target_zooms)

[oesteban]

@fliem - do you think viable putting together a PR for this?

[fliem]

@oesteban will do

[effigies]

Cool, thanks. Sorry about dropping this. I was drafting something to see if FreeSurfer wanted to relax their tolerances instead, and then my schedule filled up.