Fmriprep 1.1.1 anat-only: T1 to MNI poor quality nonlinear registration

Hi there,

This topic is a continuation of discussions started in Fmriprep 1.1.1 anat-only: troubleshooting talairach_afd error. The original problem with freesurfer was solved, but on closer look at the automatically generated fmriprep reports, we observed that the quality of the subject <–> MNI template registration was poor.

I can also confirm that the problem of nonlinear registration is observed across all 30 OAS1 subjects processed through fmriprep 1.1.1 so far.

Error logs are included as part of the previous thread (link above).

Let me know if any further clarification is required.

Appreciate the help!
jon

The coregistration looks bad, because you used the --debug flag. It switches the coregistration parameters to less precise ones (we call them ‘testing’ which you can see in the logs), that run faster which helps debugging. We are considering changing the name of this flag in the future to avoid such confusion.

1 Like

Okay thanks, Chris. Will report back after rerunning minus the --debug flag.

Much better. Thanks, @ChrisGorgolewski!

It turns out the --debug flag was included as a default option in the wrapper scripts for our bids-app singularity containers. We had misinterpreted it as some sort of --verbose like option. Committed the changes here: https://github.com/khanlab/neuroglia-helpers/blob/master/bids-apps.tsv

Also, in case you’re interested, the T1<–>MNI svg is available here.

Just went through all 30 generated reports. The quality of nonlinear registration to the template is much better.

One suggestion for future versions of fmriprep is to make it really obvious to the user that they are in DEBUG mode in the generated report, perhaps along with a disclaimer/warning that the generated outputs as a result may be suboptimal.

Happy to submit this as an enhancement/issue if you think this could be of benefit.

1 Like

A pull request on GitHub with this contribution would be absolutely fantastic. I already opened an issue to track it https://github.com/poldracklab/fmriprep/issues/1213

1 Like

MANAGED BY INCF