Fall-back registration for BBR failures

[effigies]

Original report: https://neurostars.org/t/strange-epi-to-t1w-registration-result/696

Summary: In cases of low gray/white tissue contrast in BOLD series, the default, 9-dof boundary-based registration (BBR) to T1w space can result in noticeable failures. Using the initial transform (vanilla FLIRT for FSL, mri_coreg for bbregister) can reduce the distortion.

cc @chrisfilo @oesteban @jdkent

[effigies]

@chrisfilo wrote:

do you think there would be a reliable way to automatically decided based on the input data whether or not to use BBR on a case by case basis?

If we operate on the assumption that the first pass will be pretty good, and BBR should only induce minor tweaks, we could set a number of thresholds on things like:

1. Difference in translation from initial transform to BBR transform
2. Difference in scale factors from init to BBR
3. Max BBR cost

If the threshold is passed, we could simply grab the initial transform.

[chrisgorgo]

Another option is to look at the difference in overlap of T1w and BOLD masks before and after BBR.

…

On Thu, Aug 24, 2017 at 5:39 AM, Chris Markiewicz ***@***.***> wrote: @chrisfilo <https://github.com/chrisfilo> wrote <https://neurostars.org/t/strange-epi-to-t1w-registration-result/696/8>: do you think there would be a reliable way to automatically decided based on the input data whether or not to use BBR on a case by case basis? If we operate on the assumption that the first pass will be pretty good, and BBR should only induce minor tweaks, we could set a number of thresholds on things like: 1. Difference in translation from initial transform to BBR transform 2. Difference in scale factors from init to BBR 3. Max BBR cost If the threshold is passed, we could simply grab the initial transform. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#681 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAOkpx3Lq9MhPntXZBHTizxDc_iLMaEPks5sbJwZgaJpZM4PAnQP> .

[effigies]

Okay, digging into bbregister a bit, I think the easiest way to do this will be to split out the mri_coreg pre-registration step, rather than try to recover the intermediate values. While we do get the intermediate registration transforms, we don't get the files needed for generating reports. So my thought is to create an MRICoreg interface in nipype and MRICoregRPT in niworkflows.

This would make the pattern in bbreg_wf and fsl_bbr_wf more similar, and make creating a criterion node to select the better registration (similar to that in nonlinear_sdc_wf) more straightforward.

[effigies]

I'm assuming this is something we're trying to get in for 1.0.0?

[chrisgorgo]

I think it would be worthwhile. Even though this is the first time we are seeing such problems, I have a sense it will happen more often (especially with legacy data).

…

On Thu, Aug 24, 2017 at 9:05 AM, Chris Markiewicz ***@***.***> wrote: I'm assuming this is something we're trying to get in for 1.0.0? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#681 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAOkp3oIGUAdjJEfoZ2V47YUpCEO5xxKks5sbZ9ngaJpZM4PAnQP> .

[effigies]

This issue has been replicated in ds000108, as well.

[oesteban]

Waiting for #743 to get in since there'll be conflicts.