BAMF - NNUnet Lung and Nodules

vanossj commented 6 months ago

@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 commented 6 months ago

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

jithenece commented 6 months ago

@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 commented 6 months ago

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 commented 6 months ago

Yes, that is the sct code we used

LennyN95 commented 6 months ago

Added NODULE to segdb (920c7f3).

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

LennyN95 commented 6 months ago

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 https://github.com/MHubAI/mhubio/commit/3eb4529df84a9abf0f429cba34e42c9a35a4b00b. 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 commented 5 months ago

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 commented 5 months ago

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

jithenece commented 4 months ago

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 commented 4 months ago

Archive.zip Slicer screenshots attached

jithenece commented 3 months ago

@LennyN95 Let me know if you need other changes.

jithenece commented 3 months ago

/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 commented 3 months ago

/review

@LennyN95 all review comments have been fixed

LennyN95 commented 3 months ago

NOTE: New model version already available in https://github.com/MHubAI/models/pull/92.

jithenece commented 3 months ago

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

MHubAI / models

BAMF - NNUnet Lung and Nodules #78