Improving MS lesion ground segmentation

[jcohenadad]

Improvement from @Nilser3 (rater2): https://amubox.univ-amu.fr/s/7gF5EqynwG3CkT6

Feedback from @jcohenadad (only reviewed MRS00 --> MRS11): feedback_seg_nilser.zip

Overall, I find the segmentations too "spotty". See guidelines for Basel, and notably this comment.

[Nilser3]

Hi @jcohenadad

I have followed your feedback and the guidelines,
Here is the QC for the subjects MRS00 to MRS11

[jcohenadad]

This is much better Nilser!

Minor comments:

sub-MRS09_ses-M0_UNIT1.nii.gz

• I would remove the segmentation above C1:
  

sub-MRS09_ses-M24_UNIT1.nii.gz

• I would remove the segmentation above C1
• I find the lesions slightly under-segmented
  

[Nilser3]

Following the previous comments,
here the improved MS lesion GT for marseille-3T-mp2rage

[jcohenadad]

It looks good @Nilser3 , good job 👏
These modified GTs can be added to the database. What is the strategy here: to replace rater1 or to add rater2?
And if we add rater2, how can we be sure that in the future, we will use the correct GTs for training models?

[Nilser3]

Thanks you @jcohenadad ,
I think it would be good to add the rater2 masks to the database and explain on the intranet that it is necessary to read the .json files to see the difference between masks.

Actually:

cat sub-MRS00_ses-M0_UNIT1_label-lesion_rater1.json

{
  "Author": "Sarah Demortière",
  "Date": "2019-01-01"
}

cat sub-MRS00_ses-M0_UNIT1_label-lesion_rater2.json

{
  "Author": "Sarah Demortière",
  "Date": "2019-01-01"
}

{
  "Author": "Nilser Laines Medina",
  "Date": "2023-09-01"
  "Note": "After consensus on the number and location of lesions"
}

[Nilser3]

MS ground truth improvements on:

branch: nlm/ms_lesion_improvement
commit 2b628b8c86fb956937da90be17615643c6d49d8b

ready for PR

[Nilser3]

Hello @mguaypaq
Could you please check my PR?

thanks

[mguaypaq]

This is only changing the *_label-lesion_rater2.nii.gz image files, but not the associated JSON sidecars, is this normal? Shouldn't there be a new step listed in the JSON, or at least a more recent date? Presumably these improved files are from after 2023-09-01.

Also, I'll note that the JSON sidecars are not properly formatted, because they contain 2 top-level json objects, instead of a single one, and there's a missing comma. I know we're still discussing the exact details in #282, but at least this would be a reasonable syntax:

{
  "GeneratedBy": [
    {
      "Author": "Sarah Demortière",
      "Date": "2019-01-01"
    },
    {
      "Author": "Nilser Laines Medina",
      "Date": "2023-09-01",
      "Note": "After consensus on the number and location of lesions"
    }
  ]
}

[Nilser3]

Thanks for the comments @mguaypaq
you are right, I updated the date and I have changed the associated JSON sidecars for *_label-lesion_rater2.nii.gz files.

branch: nlm/ms_lesion_improvement
commit: 2a5d27389d65853924bafcd1f7ccc0f5d73ee9b5

ready for PR

[mguaypaq]

Happy new year!

I looked at your most recent commit, and the JSON files look good now, thanks. But I'd like to check: your commit message says add non-den images, but the actual commit only touches the JSON files, is this normal? Did you also want to update the images themselves?

Also, I think the _rater1/_rater2 suffix is not really BIDS, maybe it could go in a _desc-rater1_ entity instead, but that's a separate issue, and it can be a separate pull request.

[Nilser3]

Thank you @mguaypaq , happy new year to you too 👍

I looked at your most recent commit, and the JSON files look good now, thanks. But I'd like to check: your commit message says add non-den images, but the actual commit only touches the JSON files, is this normal? Did you also want to update the images themselves?

my bad, I didn't write a good message

Also, I think the _rater1/_rater2 suffix is not really BIDS, maybe it could go in a _desc-rater1_ entity instead, but that's a separate issue, and it can be a separate pull request.

Indeed, I have renamed all the rater1/rater2 files (bids-validator seems to be happy)

Latest modifications :

branch: nlm/ms_lesion_improvement
commit: eb639675a5a5ae2bcd722e8f03d8c85e3cd24d88

ready for PR

[mguaypaq]

Ok, I reworded the commit message (so the commit is now 3b3ebce55565a775fb1d5946f0a16e3027528324), and merged into master. Thanks!

[jcohenadad]

Reopening, because it would be important to clarify in the README what those manual segmentations represent. Otherwise, a student in the future will consider rater1 and rater2 as two independent annotations, which would be wrong.

suggestion:

## derivatives

- sub-MRSXX_ses-MYY_UNIT1_label-SC_seg
- sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater1: done by XXX
- sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater2: done by XXX, and corrected by Nilser under the supervision of Julien. We recommend to ONLY use these annotations.

[Nilser3]

done @jcohenadad
branch: nlm/update_readme
commit: 4940740eedb977b866ef26fa88e276584a50de27

ready for PR @mguaypaq

[mguaypaq]

The change looks good, merged into master:

commit 4940740eedb977b866ef26fa88e276584a50de27 (HEAD -> master, origin/master, origin/HEAD)
Author: Nilser Laines Medina <[email protected]>
Date:   Tue Jan 23 09:42:42 2024 -0500

    Update README file, adding description of the raters and manual segmentations

diff --git a/README.md b/README.md
index 90f63ff..ee6a4e8 100644
--- a/README.md
+++ b/README.md
@@ -27,6 +27,12 @@ sub-MRSXX_ses-MYY_UNIT1
 - sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater1
 - sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater2
 
+## derivatives
+
+- sub-MRSXX_ses-MYY_UNIT1_label-SC_seg : done by Samira Mchinda, from SC masks generated by sct_deepseg_sc, SCT Version 5.3.0
+- sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater1: done by Sarah Demortière
+- sub-MRSXX_ses-MYY_UNIT1_label-lesion_rater2: done by Sarah Demortière, and corrected by Nilser under the supervision of Julien (see https://github.com/neuropoly/data-management/issues/267). We recommend to ONLY use these annotations.
+
 ## Authors: 
 
 - Virginie Callot,