BAMF - NNUnet Lung and Nodules

[jithenece]

No description provided.

[vanossj]

@LennyN95 we could use your advise on this model. We run two different nnunet models, then combine the output niftis into one nifti, then convert to dicom-seg.

• whats the best way to run multiple nnunet models? seems like you can only have one instance of NNUnetRunner module.
• how does the ROI information get passed through if there are multiple model runner modules?

[LennyN95]

Hi @vanossj,

Sure, that is possible! So, the module config is the general configuration for all usages of that module. However, that can be overwritten directly from within the execution chain:

execute:
- module: NNUnetRunner
  nnunet_task: Task003_Liver
  roi: LIVER,LIVER+NEOPLASM_MALIGNANT_PRIMARY
- module: NNUnetRunner
  nnunet_task: Task007_Pancreas
  roi: PANCREAS,PANCREAS+NEOPLASM_MALIGNANT_PRIMARY

modules:
  NNUnetRunner:
    nnunet_model: 3d_lowres

[LennyN95]

To avoid file overwrites, the @IO.Output decorator has an auto_increment option:
https://github.com/MHubAI/mhubio/blob/main/mhubio/core/IO.py#L325

[jithenece]

@LennyN95

• Do we need to update this NNUnetRunner function to process this auto_increment option?
• segdb doesnot have entries for LUNG, NODULES. Could you add this for us?

[LennyN95]

Do we need to update this NNUnetRunner function to process this auto_increment option?

That is correct. I will take care of this and release a new base image including this and some additional minor fixes.

segdb doesnot have entries for LUNG, NODULES. Could you add this for us?

So you need NODULE with sct code 27925004, correct?

[vanossj]

Yes, that is the sct code we used

[LennyN95]

Added NODULE to segdb (920c7f3).

You can now use LUNG+NODULE or LEFT_LOWER_LUNG_LOBE+NODULE, etc.

[LennyN95]

Do we need to update this NNUnetRunner function to process this auto_increment option?

The default of the auto_increment option in the @IO.Output and @IO.Outputs decorators has been changed to True with MHubAI/mhubio@3eb4529. Hence, when running multiple NNUnetRunner modules, previous outputs shouldn't be overwritten.

These changes are available in our latest base image mhubai/base:latest.

[jithenece]

Added NODULE to segdb (920c7f3).

You can now use LUNG+NODULE or LEFT_LOWER_LUNG_LOBE+NODULE, etc.

I tried to run testing with the LUNG,LUNG+NODULE as segmentation covers both right and left lung but got error. Could you add LUNG to segdb too?

[LennyN95]

I also added LUNG now (we already had the type in the DB but not the segment).

[jithenece]

sample:
  idc_version: "Data Release 4.0 May 29, 2020"
  data:
  - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.7777.9002.290917407640793598352855735724
    aws_url: s3://idc-open-data/b925cf2a-de89-426d-b300-27b136264555/*
    path: input_data

reference:
  url: https://drive.google.com/file/d/1la5-sFlaWGopGiBvTdi-_V375nc59e9h/view?usp=sharing

[jithenece]

Archive.zip
Slicer screenshots attached

[LennyN95]

Very clean implementation! I added a few questions and minor remarks. Thank you!

[LennyN95]

It seems that bamf_nnunet_ct_lungnodules would be a better fit for a name?

[jithenece]

I have updated with this naming convention.

[LennyN95]

thx! This would require to change the folder name (modles/$model_name) too.

[jithenece]

updated the model directory to bamf_nnunet_ct_lungnodules

[jithenece]

@LennyN95 Let me know if you need other changes.

[jithenece]

/test

sample:
  idc_version: "Data Release 4.0 May 29, 2020"
  data:
  - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.7777.9002.290917407640793598352855735724
    aws_url: s3://idc-open-data/b925cf2a-de89-426d-b300-27b136264555/*
    path: input_data

reference:
  url: https://drive.google.com/file/d/1pTqmrrQWC4qHO8z4d72FPVAczh-Zj7hq

[jithenece]

/review

@LennyN95 all review comments have been fixed

[LennyN95]

NOTE: New model version already available in #92.

[jithenece]

@LennyN95 Closing this pull request as new model #92 to be used as per the discussion in comments here