Cropped edges in registration to the template

[jcohenadad]

When working on #31 I noticed cropped edges in the registered template:

I suspect this is caused by improper straightening (too large curvature?). I'll investigate

[jcohenadad]

The segmentation goes to the last bottom slice:

And the straightening looks OK:

The NURBS-based fitting looks OK:

[jcohenadad]

The straight2curve warping field applied to the straightened image does not result in the same original image (notice the discs move a lot, and the cord seems slightly larger on the registered image):

Although I don't think this explains the multiple slices that are missing on the edges, this is highly concerning and an issue should be opened on SCT

EDIT 20240514_094659: Issue opened: https://github.com/spinalcordtoolbox/spinalcordtoolbox/issues/4477

[jcohenadad]

@Kaonashi22 now that https://github.com/spinalcordtoolbox/spinalcordtoolbox/issues/4477 is solved, I'm rerunning processing on 14 subjects to see if this issue is resolved on your data. I'll report results here.

[Kaonashi22]

Thanks @jcohenadad!

[jcohenadad]

OK, in general, the fix improves results a lot. However, some issues still remain, that I'm investigating now. For example, with subject sub-DEV169Sujet03, the fix still produced cropped edges.

Before fix:

After fix:

[jcohenadad]

Comparing the input image and the input-straight-curved shows highly similar content overall (thanks to https://github.com/spinalcordtoolbox/spinalcordtoolbox/pull/4514), except at the top/bottom edge:

Maybe this is the cause of the problem?

It is surprising, given that many slices are affected at the bottom (about 30), even though the warping field at the bottom seems only affected over a few slices, and the segmentation is fine towards the bottom:

[jcohenadad]

The first landmark-based rigid registration did not work:

Leading to a strong cropping of the straightened image:

[jcohenadad]

Going into debug mode...

The input label files to step=0 are properly positioned and oriented:

[jcohenadad]

Checking on another subject where this cropping is not/less apparent: sub-BB277

Ah! Still not working properly:

[jcohenadad]

Has this issue been introduced by https://github.com/spinalcordtoolbox/spinalcordtoolbox/pull/4514? Let's re-run SCT before the fix...

nope! Same problem before the fix:

[jcohenadad]

@Kaonashi22 good news, the issue has been identified and fixed in https://github.com/spinalcordtoolbox/spinalcordtoolbox/pull/4533. Once @joshuacwnewton reviews it we will merge it and you can resume your experiments.

[Kaonashi22]

Thanks everyone!!!

[jcohenadad]

@Kaonashi22 to be 100% sure I'm also re-running the processing on the 14 subjects you shared with me. I'll share the QC report once done.

EDIT 20240618_142034:

• QC: https://www.dropbox.com/s/r1ai2cgpnkc4u32/20240618_qc_lydia.zip?dl=0
• results: results.zip
• log: log.zip

Two crashes:

• https://github.com/sct-pipeline/spine-park/issues/36
• https://github.com/sct-pipeline/spine-park/issues/37

[joshuacwnewton]

Sorry for the delay! The PR has been merged. Feel free to update your local copies of SCT and retry.

[jcohenadad]

@Kaonashi22 you need to do the following:

cd $SCT_DIR
git pull

And then you can re-run the processing on all subjects. For every new issue you experience, please open a specific issue. I'm closing this issue now.