Replace of seg_ms_lesion_mp2rage by UNIseg model on sct_deepseg

[Nilser3]

Description

This PR replace the current seg_ms_lesion_mp2rage model by UNIseg (single-class 3D nnUnet model discused here #75 ) keeping the same name: seg_ms_lesion_mp2rage (after comparisons these models see #81 )
Updates also the README file, explaining how to use the model, datasets for training and acknowledgments to the collaborators.

[jcohenadad]

@Nilser3 indicate where to find model to test

[Nilser3]

@Nilser3 indicate where to find model to test

Thanks you @jcohenadad
The model is here https://github.com/ivadomed/model_seg_ms_mp2rage/releases/tag/r20240610

I have already replaced the seg_ms_lesion_mp2rage model and it is ready to be tested this branch:

branch on SCT: nlm/update-_seg_ms_lesion_mp2rage-model
commit: b64340b4259aa99734a4ef809edf4af882261409

[jcohenadad]

Getting this message during inference:

/Users/julien/code/spinalcordtoolbox/python/envs/venv_sct/lib/python3.9/site-packages/nnunetv2/utilities/plans_handling/plans_handler.py:37: UserWarning: Detected old nnU-Net plans format. Attempting to reconstruct network architecture parameters. If this fails, rerun nnUNetv2_plan_experiment for your dataset. If you use a custom architecture, please downgrade nnU-Net to the version you implemented this or update your implementation + plans.
  warnings.warn("Detected old nnU-Net plans format. Attempting to reconstruct network architecture "

any concerns?

[jcohenadad]

Another message, possibly concerning:

Found Intel OpenMP ('libiomp') and LLVM OpenMP ('libomp') loaded at
the same time. Both libraries are known to be incompatible and this
can cause random crashes or deadlocks on Linux when loaded in the
same Python program.
Using threadpoolctl may cause crashes or deadlocks. For more
information and possible workarounds, please see
    https://github.com/joblib/threadpoolctl/blob/master/multiple_openmp.md

[jcohenadad]

The zipped model does not seem to include the list of subjects used for training and the version of the dataset used to train the model. It is required.

[jcohenadad]

Inference time is pretty long... I've tried without cropping on the basel data, inference time was 4min 6s.

With cropping: 7s 😊

[jcohenadad]

Comparison with previous model, and current model with various crop sizes:

Conclusion: too narrow cropping (5x5x5) leads to bad segmentation. A minimum cropping size is required. What size? @Nilser3 that's a follow-up investigation

[jcohenadad]

Conclusion: too narrow cropping (5x5x5) leads to bad segmentation. A minimum cropping size is required. What size? @Nilser3 that's a follow-up investigation

follow-up idea: given that the inference is sliiiiightly different across crop sizes, how about computing a soft segmentation based on the inference performed across various crop sizes?

[Nilser3]

The zipped model does not seem to include the list of subjects used for training and the version of the dataset used to train the model. It is required.

Thank you @jcohenadad,
Ok, the release r20240610 has been updated with the needed information!

[jcohenadad]

@Nilser3 any update? this is becoming urgent: https://forum.spinalcordmri.org/t/segmentation-of-spinal-cord-lesion-on-mp2rage-uni-images-doesnt-work/1302

[Nilser3]

@Nilser3 any update? this is becoming urgent: https://forum.spinalcordmri.org/t/segmentation-of-spinal-cord-lesion-on-mp2rage-uni-images-doesnt-work/1302

The model was replaced in the release 6.4, this branch is ok for PR.