Add feature to use manually corrected labels on T2w data

[jcohenadad]

As per #15, the script should be modified to use manually corrected labels if they exist. The labels should be located under ./derivatives/labels/ and have the suffix _seg on the T2w data.

[jcohenadad]

Ah! It's already implemented 😅

And it is working. I tested it with a derivatives/labels T2w segmentation mask in one subjects, and the script used it:

👉 Processing: sub-BB277_T2

Looking for manual segmentation: /Users/julien/temp/Lydia/data_small/derivatives/labels/sub-BB277/anat/sub-BB277_T2_seg.nii.gz
Found! Using manual segmentation.

Closing

[jcohenadad]

@Kaonashi22 you can already use the manually corrected segmentations for your project.

[Kaonashi22]

Version of the script: 78129ca

Regarding the issue about segmentations being overwritten, here is the path to the input directory: /dagher/dagher12/lydia12/SPINE_park/CENIR_ICEBERG_BIDS. There is no "derivatives" folder. I don't think the script generates this folder (and moves the segmentations there).

[jcohenadad]

There is no "derivatives" folder.

Ah, so that's the problem. The manual segmentations should be located under a specific derivatives/ folder as per: #17 (comment)

I'll update the README with a clear example--

[Kaonashi22]

So, I should create the derivatives folder myself and move the segmentation files there?

[jcohenadad]

yes, or, much better, use an automatic procedure that also creates the JSON file (important for tracking provenance): https://github.com/spinalcordtoolbox/manual-correction

we covered this in the latest SCT course (maybe @valosekj you know the exact passage in the video/slides)

[valosekj]

we covered this in the latest SCT course (maybe @valosekj you know the exact passage in the video/slides)

Yes! The passage about manual corrections starts at 6:44:26.
We also have a step-by-step example of SC segmentation corrections on our wiki.
There is also this YT tutorial on correcting segmentations across multiple subjects. However, it is slightly obsolete (we now use https://github.com/spinalcordtoolbox/manual-correction instead of the script described in the video). But the overall concept is the same.

Let me know if you need any further info; I will be happy to help!

[Kaonashi22]

Thanks @valosekj and @jcohenadad.
I actually used the "manual-correction" module to correct the masks. In the end, the derivatives folder only includes subjects with these corrected masks, not those that didn't required a manual correction (as they were not in the yml file). So their segmentations are overwritten.
I guess I should manually export the masks of these subjects to the derivatives folder as well.

[Kaonashi22]

Would you know if the segmentations are identical between different runs? I already did the QC and manual corrections on previous segmentations. After reruning the pipeline, I obtained new segmentations, and I'm wondering whether I need to redo the QC and corrections.

[jcohenadad]

I actually used the "manual-correction" module to correct the masks. In the end, the derivatives folder only includes subjects with these corrected masks, not those that didn't required a manual correction (as they were not in the yml file). So their segmentations are overwritten.
I guess I should manually export the masks of these subjects to the derivatives folder as well.

No you should not. What you describe is the correct procedure, except that the output folder should be a pristine folder, to avoid any conflict with previously-processed data. The way I do it, is by setting -path-output in my sct_run_batch command to be uniquely identified by the date and time, eg: -path-output results_20240411_091115. That way, the non-manually corrected segmentations are not overwritten, but are simply re-created. This is the best way to ensure reproducibility of the whole pipeline.

If you run out of space, feel free to delete your old results_* folders.

I hope that's clearer.

Would you know if the segmentations are identical between different runs?

Yes they are identical. Hence the concept of reproducibility. The only thing you need to care about are the manual corrections, which you should put under the derivatives folder.

[Kaonashi22]

I see, that's clearer, thanks!

[valosekj]

In the end, the derivatives folder only includes subjects with these corrected masks

Yes, this is correct! The manual-correction module saves only the manually corrected masks under derivatives.

, not those that didn't required a manual correction (as they were not in the yml file).

Just for info. If you really needed to save masks that don't require manual correction under derivatives, you could you the -add-seg-only argument (described here). This argument will copy masks that aren't in -config list (i.e., those that didn't require correction) into derivatives.
This might be useful, for example, if you wanted to store masks generated by a specific algorithm version.

[Kaonashi22]

Sounds good, thanks for these explanations!

…

________________________________ From: Jan Valosek ***@***.***> Sent: April 11, 2024 21:14 To: sct-pipeline/spine-park ***@***.***> Cc: Lydia Chougar, Dr ***@***.***>; Mention ***@***.***> Subject: Re: [sct-pipeline/spine-park] Add feature to use manually corrected labels on T2w data (Issue #17) In the end, the derivatives folder only includes subjects with these corrected masks Yes, this is correct! The manual-correction<https://github.com/spinalcordtoolbox/manual-correction> module saves only the manually corrected masks under derivatives. , not those that didn't required a manual correction (as they were not in the yml file). Just for info. If you really needed to save masks that don't require manual correction under derivatives, you could you the -add-seg-only argument (described here<https://github.com/spinalcordtoolbox/manual-correction/blob/main/manual_correction.py#L230-L235>). This argument will copy masks that aren't in -config list (i.e., those that didn't require correction) into derivatives. This might be useful, for example, if you wanted to store masks generated by a specific algorithm version. — Reply to this email directly, view it on GitHub<#17 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BFCFJYVSIYVC34LFEOIB4ELY44YO5AVCNFSM6AAAAABFYAOL4GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJQG44TIMRVGE>. You are receiving this because you were mentioned.Message ID: ***@***.***>