BAMF - NNUnet Kidney CT

[rahul99]

No description provided.

[LennyN95]

@rahul99 didn't this model delineate kidney cysts too?
We're about to add an entry to SegDB for kidney cysts after consultation with @dclunie.

[LennyN95]

I extended our SegDB with the morphologically abnormal structure CYST.

Now, you can use LEFT_KIDNEY+CYST to encode a cyst in the left kidney or KIDNEY+CYST in case your nnUNet algorithm does not distinguish between cysts in the left or right kidney.

[fedorov]

@rahul99 for testing the model, please provide the following for a sample CT series that is known to produce acceptable results with this model:

*SeriesInstanceUID

• aws_url for the bucket folder containing the series files
• version of IDC data that you used for getting the above

If you pick one of the cases that are already accompanied by a segmentation produced by this model in IDC (they have just been added yesterday in IDC v17), you can also include the items above for the SEG series.

[rahul99]

/test

sample:
  idcv: 17.0
  data:
    - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795
      aws: 43b672b6-dd41-4a00-a1ad-dd5a8cf407df
      path: dicom

reference:
  url: https://www.dropbox.com/scl/fi/v517ibmkdggsz3kgmfd9g/bamf_nnunet_ct_kidney.zip?rlkey=4ecm7zd23g1vqwh444dytyxen&dl=0

[fedorov]

aws: 43b672b6-dd41-4a00-a1ad-dd5a8cf407df

@rahul99 I believe you need to update this to follow the updated schema (attribute names changed, full S3 URL is required), see here for the example: #71 (comment).

[rahul99]

/test

sample:
  idc_version: 17.0
  data:
    - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795
      aws_url: s3://43b672b6-dd41-4a00-a1ad-dd5a8cf407df/*
      path: dicom

reference:
  url: https://www.dropbox.com/scl/fi/v517ibmkdggsz3kgmfd9g/bamf_nnunet_ct_kidney.zip?rlkey=4ecm7zd23g1vqwh444dytyxen&dl=0

[fedorov]

aws_url: s3://43b672b6-dd41-4a00-a1ad-dd5a8cf407df/*

This is not a valid URL. To look up the URL you can use the following query:

SELECT
  DISTINCT(series_aws_url)
FROM
  `bigquery-public-data.idc_current.dicom_all`
WHERE
  SeriesInstanceUID = '1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795'

Which returns s3://idc-open-data/43b672b6-dd41-4a00-a1ad-dd5a8cf407df/.

[fedorov]

/test

sample:
  idc_version: 17.0
  data:
    - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795
      aws_url: s3://idc-open-data/43b672b6-dd41-4a00-a1ad-dd5a8cf407df/*
      path: dicom

reference:
  url: https://www.dropbox.com/scl/fi/v517ibmkdggsz3kgmfd9g/bamf_nnunet_ct_kidney.zip?rlkey=4ecm7zd23g1vqwh444dytyxen&dl=0

[rahul99]

Thanks @fedorov , I was just verifying the url and was about to update it. I see you already did that, thanks :)

[rahul99]

/test

sample:
  idc_version: 17.0
  data:
  - SeriesInstanceUID: 1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795
    aws_url: s3://idc-open-data/43b672b6-dd41-4a00-a1ad-dd5a8cf407df/*
    path: dicom

reference:
  url: https://www.dropbox.com/scl/fi/v517ibmkdggsz3kgmfd9g/bamf_nnunet_ct_kidney.zip?rlkey=4ecm7zd23g1vqwh444dytyxen&dl=0

---

Test Results (24.03.28_17.30.31_kb1YE6N1mX)

id: b557241e-1931-4ce4-b58e-f8f408baf242
date: '2024-03-28 17:46:32'
checked_files:
- file: bamf_nnunet_ct_kidney.seg.dcm
  path: /app/test/src/1.3.6.1.4.1.14519.5.2.1.9203.4004.385483163880959624431394063795/bamf_nnunet_ct_kidney.seg.dcm
  checks:
  - checker: DicomsegContentCheck
    notes:
    - label: Segment Count
      description: The number of segments identified in the inspected dicomseg file.
      info: 3
summary:
  files_missing: 0
  files_extra: 0
  checks:
    DicomsegContentCheck:
      files: 1
conclusion: true

[LennyN95]

@rahul99 @fedorov You can just edit a test request comment to correct any mistakes. A new test request can be made if data changed or the model needs to be tested again because the implementation changed.

Note, that the code block must be yaml (thus ```yaml [..] ```)

[LennyN95]

Is there any additional information available about the demographic / clinical history of the training patients?

[vanossj]

Information about the training data is in the 'training' section, that has links to the datasets used for training which has the demographic info

[LennyN95]

That's fine, I suggest we add a short notice then here too (maybe just refer to the test section).

In the worst case, a reader might not even continue to the test section.

[rahul99]

Hey @LennyN95 , I have updated meta with whatever information I could find in the source paper. Do I need to comment the test case again or will the last one be picked up automatically since there is no change in the test?

[LennyN95]

Perfect! The passed test is still valid then, you won't need to redo it. I'll review in a bit and then continue to merge and push. Great work @rahul99, it's so great to see the platform growing now!

[LennyN95]

@rahul99 could you upload 1-5 images for the model card that show the model in action? This is not a hard requirement, thus I moved on with the push. However, from a UX perspective, visuals add a great value to the model page and also help readers to quickly recognize a model (especially as the model card is made to look similar across all models, images can be a great help here).