BAMF NNUnet Lung and Nodules (v2)

[jithenece]

Pretrained model for 3D semantic image segmentation of the lung and lung nodules from ct scan

[jithenece]

/test

sample:
  idc_version: "Data Release 3.0 September 24, 2021"
  data:
  - SeriesInstanceUID: 1.2.840.113654.2.55.302533635268177260202698156939540188162
    aws_url: s3://idc-open-data/f038cf01-a72a-44a7-88a1-d25d18f08ddf/*
    path: input_data

reference:
  url: https://drive.google.com/file/d/1j6VFntIlsssnzFlbOW1sziBwbpYeN_Ah/view?usp=sharing

[LennyN95]

@jithenece Is this a separate version? It looks like it's the same model as in #78 if that is the case, please implement all changes in that PR (it's not yet merged).

[jithenece]

@jithenece Is this a separate version? It looks like it's the same model as in #78 if that is the case, please implement all changes in that PR (it's not yet merged).

@LennyN95 PR #78 has been implemented as part of AIMI1 and it uses 2 different models to finally generate the lung and nodule segmentation. This PR #92 is part of AIMI2 and uses a single model to generate the same output but has better results. Hence I had raised this PR by incrementing version as 2.0.0. Please let me know your thoughts.

[jithenece]

Both are listed in the zenodo

• lung-ct.zip
• lung2-ct.zip

[LennyN95]

Great! Since we didn't freeze the v1 model yet as part of a MHub release (in fact, the other model is in testing stage and yet to be merged), this would effectively overwrite the v1 implementation. Do you think there is value to have v1 and v2 coexist on MHub? If yes, I suggest to implement this model as a separate model into MHub with a _v2 suffix to the model name.

[vanossj]

I think its ok to close #78 and have this model as the only bamf lung nodule. The model in #78 can be run from a docker container as described in the repo, so if anyone whats to reproduce our results they can, even if its not on mhub.

[LennyN95]

Review is done with some minor comments (many overlap with those in the other PRs).

[LennyN95]

Just to confirm, the model suports CT scans with a slice thickness up to 1cm?

[jithenece]

Adding segmentation file to generate permanent link for input below
output.zip

` idc_version: 2.0
data:

• SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.4320.5030.194126952887269510493034552293
  aws_url: s3://idc-open-data/cea2aa44-31aa-40ff-95ab-020baae065ff/*
  path: input_data
  `

[jithenece]

/test

sample:
  idc_version: 2.0
  data:
  - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.4320.5030.194126952887269510493034552293
    aws_url: s3://idc-open-data/cea2aa44-31aa-40ff-95ab-020baae065ff/*
    path: input_data

reference:
  url: https://github.com/user-attachments/files/16761594/output.zip

---

Test Results (24.08.29_15.15.30_AkoraywFNq)

id: e878cb29-fae1-49ae-aec2-6a8510c13eb2
date: '2024-08-29 15:59:21'
missing_files:
- 1.3.6.1.4.1.14519.5.2.1.4320.5030.194126952887269510493034552293/bamf_nnunet_ct_lungnodules.seg.dcm
summary:
  files_missing: 1
  files_extra: 0
  checks: {}
conclusion: false

[jithenece]

/test

sample:
  idc_version: 2.0
  data:
  - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.4320.5030.194126952887269510493034552293
    aws_url: s3://idc-open-data/cea2aa44-31aa-40ff-95ab-020baae065ff/*
    path: input_data

reference:
  url: https://github.com/user-attachments/files/16761594/output.zip

Test Results (24.08.29_15.15.30_AkoraywFNq)

Is it possible to get more verbose logs from the docker?

[LennyN95]

Please note, we updated our base image. All mhub dependencies are now installed in a virtual environment under /app/.venv running Python 3.11. Python, virtual environment and dependencies are now managed with uv. If required, you can create custom virtual environments, e.g., uv venv -p 3.8 .venv38 and use uv pip install -p .venv38 packge-name to install dependencies and uv run -p .venv3.8 python script.py to run a python script.

We also simplified our test routine. Sample and reference data now have to be uploaded to Zenodo and provided in a mhub.tom file at the project root. The process how to create and provide these sample data is explained in the updated testing phase article of our documentation. Under doi.org/10.5281/zenodo.13785615 we provide sample data as a reference.

[jithenece]

@LennyN95 I have updated this and other pending models too. Please let me know if any changes required.

[LennyN95]

Test Results

id: 405eb452-2a80-494b-b341-40502d135cd3
name: MHub Test Report (default)
date: '2024-10-09 11:18:27'
missing_files:
- 1.3.6.1.4.1.14519.5.2.1.4320.5030.194126952887269510493034552293/bamf_nnunet_ct_lungnodules.seg.dcm
summary:
  files_missing: 1
  files_extra: 0
  checks: {}
conclusion: false

[LennyN95]

The NNUnetRunnerV2 module fails with FileNotFoundError: [Errno 2] No such file or directory: '/app/tmp/9402fb71-3026-4da8-9c19-f660e64f5cc2/VOLUME_001.nii.gz'.

[LennyN95]

@jithenece I see you fixed the version of nnunetv2==2.0 - are there any incompatibilities? We are currently working with the nnunet maintainers on a feature to disable multiprocessing during inference (see MIC-DKFZ/nnUNet#2556). Please let me know if switching to the latest version would cause any problems.

[LennyN95]

@jithenece in addition to my previous post, @FJDorfner suggested the following setting (8ef8980) and it seems to work fine with the nnunetv2 version. Can you give us some brief feedback on how you tried this so we understand any edge cases. Thanks a lot!

[jithenece]

@LennyN95 I have set it to 2.0 as it was trained using this version. Also to avoid any issues if the newer nnunetv2 has changes breaking this code base.

I tried above flags on nnunetv2-2.5.1 to check if the newer flags can work but got below error
nnUNetv2_predict: error: unrecognized arguments: --nproc_export 1 --nproc 1 --batchsize 1

Could you share the nnunetv2 version which should be used?