Integrate baselines from CNMC_PAI (organizer baseline)

[zjiang0529]

In response to Issue #116, add baselines from CNMC_PAI for 7 tasks.

[Copilot]

Pull request overview

Adds BraTS25_B (organizer baseline) entries for seven BraTS25 task algorithms — Adult Glioma Pre/Post, Africa, Meningioma, Meningioma RT, Metastases, Pediatric, and Inpainting — plumbing them through enums, YAML metadata, README tables, and updating the rank-integrity test to recognize the new B suffix.

Changes:

• Add BraTS25_B enum members in brats/constants.py and matching YAML entries (with aparida12/brats2025:* Docker images) in seven task metadata files, including a new Baseline rank anchor in each.
• Update tests/utils/test_algorithm_config.py rank parser to map suffix B → Baseline while preserving existing tie-suffix handling.
• Add README rows and an explanatory footnote for each new baseline, plus a BraTS-PEDs 2026 dataset paper link.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
brats/constants.py	Adds BraTS25_B enum entries with docstrings to seven *Algorithms enums
brats/data/meta/adult_glioma_pre_and_post_treatment.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:gli
brats/data/meta/africa.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:ssa (requires_root)
brats/data/meta/meningioma.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:men
brats/data/meta/meningioma_rt.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:menrt (requires_root)
brats/data/meta/metastases.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:met (requires_root)
brats/data/meta/pediatric.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:peds (requires_root)
brats/data/meta/inpainting.yml	New baseline anchor and BraTS25_B entry using aparida12/brats2025:inpt
tests/utils/test_algorithm_config.py	Extends rank parsing to map suffix B to Baseline
README.md	Adds Baseline* rows and footnotes for seven task sections, plus a PEDs 2026 dataset paper link

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[neuronflow]

@Astarakee as I understand it your baseline is not integrated here?

Should we integrate these into this same PR?

Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

[zjiang0529]

@Astarakee as I understand it your baseline is not integrated here?

Should we integrate these into this same PR?

Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

I used BraTS25_B to be consistent with the format for the other Enums. We could use BraTS25_B1, B2... for Baseline 1, Baseline 2 etc.?

[Astarakee]

@Astarakee as I understand it your baseline is not integrated here?

Should we integrate these into this same PR?

Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

I will follow the same way that Zhifan applied; for GLI, METs, MENpre, and MENrt all in 2025.

[neuronflow]

@Astarakee as I understand it your baseline is not integrated here?
Should we integrate these into this same PR?
Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

I will follow the same way that Zhifan applied; for GLI, METs, MENpre, and MENrt all in 2025.

Okay, I suggest having both baselines in the same PR, so please build upon @zjiang0529 's work and extend this PR. Would that work for you?

[neuronflow]

@MarcelRosier should we download the docker images and reupload them to our repo, or keep it as it is?

[neuronflow]

@Astarakee as I understand it your baseline is not integrated here?
Should we integrate these into this same PR?
Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

I used BraTS25_B to be consistent with the format for the other Enums. We could use BraTS25_B1, B2... for Baseline 1, Baseline 2 etc.?

For me that suggests some kind of ranking between the baselines, hence I proposed A/B to differentiate from the ranked algorithms. If @Astarakee is okay with having it portrayed like that I am also fine with B1/2.

[zjiang0529]

@Astarakee as I understand it your baseline is not integrated here?
Should we integrate these into this same PR?
Probably the Enums should account for the two Baselines, maybe by going for BraTS25_Baseline_A, BraTS25_Baseline_B or is there a more elegant solution?

I used BraTS25_B to be consistent with the format for the other Enums. We could use BraTS25_B1, B2... for Baseline 1, Baseline 2 etc.?

For me that suggests some kind of ranking between the baselines, hence I proposed A/B to differentiate from the ranked algorithms. If @Astarakee is okay with having it portrayed like that I am also fine with B1/2.

Have we decided on the name yet? A/B is clearer than 1/2, but using BraTS25_Baseline_A and B could make the columns wider and have three parts. Maybe BraTS25_BaselineA/B or BraTS25_BA/BB?

[MarcelRosier]

@MarcelRosier should we download the docker images and reupload them to our repo, or keep it as it is?

we could do that for control and consistency, but it's not strictly necessary

[MarcelRosier]

Did not get the chance to actually test it, but it looks solid. Just settle on a naming convention and ensure it does not break the tests (might be the case currently if we want to support multiple baselines). Also, I would advise moving images to the brainles repo for consistency. I can support with that if we agree on that

[PaulaTBz]

@zjiang0529
I think I found a small problem in the inpainting.yml file. I ran the code you provided with BraTS25_B and got an error regarding writing permissions ("Not enough output files were created by the algorithm. Expected: 1 Got: 0."). I added the key "requires_root" under run_args in the yaml file and set it to true (as is the case for some of the other algorithms ). Now it seems to work. I never had this issue on my laptop, it came up on a remote server where I am testing the baseline. I will check the output and try with other models as well.

[PaulaTBz]

@zjiang0529
I was able to run all baseline algorithms, except for one: Meningioma RT. I always used the example glioma data from the tutorials repo (either segmentation or inpainting data, depending on the task) and it worked for all algorithms, but for Meningioma RT.
There seems to be an issue with the renaming of the data within the container. Apparently, the container expects data of naming convention "BraTS-MEN-RT-0000-1_t1c.nii.gz", i.e. with an underscore before the modality, while in reality, the data is being renamed to something else. Most likely, there should be a hyphen ("-") instead of an underscore ("_") inside the name.

I get this error:

⠋ Running inference...Container finished with an error:
Traceback (most recent call last):
File "/project/runner.py", line 190, in
batch_processor(args.input, args.output)
File "/project/runner.py", line 160, in batch_processor

infer_single(input_path, Path(output_folder))

File "/project/runner.py", line 104, in infer_single

Running inference for task: MEN-RT
Setting up model weights for task: MEN-RT
storing artifacts in tmp dir /tmp/tmp9pwx4fre
BraTS-MEN-RT-0000-1

os.rename(input_folder_raw / name/ f'{name}{key}', input_folder_raw / name/ f'{name}{val}')

FileNotFoundError: [Errno 2] No such file or directory: '/tmp/tmp9pwx4fre/inp/BraTS-MEN-RT-0000-1/BraTS-MEN-RT-0000-1_t1c.nii.gz' -> '/tmp/tmp9pwx4fre/inp/BraTS-MEN-RT-0000-1/BraTS-MEN-RT-0000-1_0000.nii.gz'

[zjiang0529]

@PaulaTBz
Hi Paula, thank you for helping with this PR, and you're right.
For the Meningioma RT (MENRT) task, our container expects the NIfTI file name to end with _t1c.nii.gz.

This is due to a naming inconsistency in the official MENRT dataset. Other tasks use -t1c.nii.gz, but MENRT does not (at least for 2024 and 2025), so we added a special condition for this task. Without it, users or challenge testing code will encounter issues when running the container on the official dataset downloaded from Synapse.

I guess your sample data file follows the correct "-" naming conversion. You could test with a MENRT file, an intentionally "wrongly" named sample file, or the file naming in the official MENRT dataset would need to be changed, which is outside of our control.

Best,
Zhifan

[PaulaTBz]

Apart from the two small issues regarding "requires_root" and the special naming convention of input data for Meningioma RT, all baselines run successfully.

[zjiang0529]

Since all changes have been approved, could you please consider validating and merging the PR? Thanks.

[MarcelRosier]

A few items still have to addressed @zjiang0529:

Must fix (blocking merge)

1. brats/data/meta/meningioma_rt.yml — BraTS25_B missing subject_modality_separator: "_"
   4 of 6 existing MENRT algorithms set this field (BraTS25_1, 25_2, 24_1, 24_3). The aparida12/brats2025:menrt container expects "_" (underscore) as the modality separator — Paula's testing confirmed this is the cause of the FileNotFoundError. Without it, the default "-" is used and the container's internal rename logic fails.

2. brats/data/meta/inpainting.yml — BraTS25_B missing requires_root: true
   The container needs root access to run. Paula confirmed adding this fixes the "Not enough output files" error.

Discuss

3. Naming convention — BraTS25_B will conflict when Astarakee's baselines are added
   If Astarakee also adds BraTS25_B to the same enum, it'll be a duplicate key. Options raised in the thread were BraTS25_BA/BB, BraTS25_BaselineA/B, etc. Would be good to settle this before the next set of baselines lands.

4. Astarakee's baselines to be added to this same PR
   As discussed above — Astarakee agreed to follow the same pattern. Should we wait for those commits before merging?

5. Docker image hosting — aparida12/brats2025:* vs reuploading to brainles/
   I advised moving them to the brainles/ org for consistency and offered to help. Not a blocker, but worth deciding.

[zjiang0529]

Thank you for pointing out these changes. If I understand correctly, the action plan is as follows. Could you confirm?

Easy fixes:

1. Add subject_modality_separator: "_" to the YAML meta file.
2. Add requires_root: true.
3. Yes, please upload the Docker images to brainless/. If you need any action from us, please let us know.

Still unclear:
3. Who should approve the naming convention? If there are no concerns, I will proceed with BraTS25_BA/BB.
4. I would be happy to help add other baselines if they need to be in the same PR. But, also would like to move this PR forward soon if we can consider keeping our baselines together in this PR and address different baselines separately.

[neuronflow]

@PaulaTBz can you please take care of the image uploads?

[neuronflow]

3. Naming convention — BraTS25_B will conflict when Astarakee's baselines are added
If Astarakee also adds BraTS25_B to the same enum, it'll be a duplicate key. Options raised in the thread were BraTS25_BA/BB, BraTS25_BaselineA/B, etc. Would be good to settle this before the next set of baselines lands.

I suggest going for BraTS25_CNMC_PAI or any other acronym you want. This way, users don't need to look up in the table which baseline is which, and no ranking is suggested.

[neuronflow]

Thanks!
@Astarakee could @zjiang0529 help you create a PR for your baseline?

[PaulaTBz]

I pushed all baseline images to the brainles namespace on docker hub

[neuronflow]

I pushed all baseline images to the brainles namespace on docker hub

cool, now the paths need to be adjusted ;)
either you do it or @zjiang0529 needs to know the paths

[PaulaTBz]

I pushed all baseline images to the brainles namespace on docker hub

cool, now the paths need to be adjusted ;) either you do it or @zjiang0529 needs to know the paths

I can do it, but I'll post the new paths here as well for reference:

brainles/brats25_ssa_baseline
brainles/brats25_inpainting_baseline
brainles/brats25_met_baseline
brainles/brats25_menrt_baseline
brainles/brats25_men_baseline
brainles/brats25_ped_baseline
brainles/brats25_gli_baseline

The tag is always latest.

I'll wait for Zhifan to make the modifications and merge his branch, then I'll update the paths in a separate PR. (I think that makes the most sense) :)

[MarcelRosier]

any progress or blockers here? @zjiang0529 @PaulaTBz

[PaulaTBz]

any progress or blockers here? @zjiang0529 @PaulaTBz

Nothing blocking the merge from my side, assuming the small issues I mentioned above and the question about the naming convention have been addressed. Once this is merged, I'll open a new PR with the updated pointers to the brainles docker hub name space for consistency :)