MHub / GC - WSI background/tissue segmentation model

MHubAI / models

Stores the MHub models dockerfiles and scripts.

MIT License

8 stars 16 forks source link

MHub / GC - WSI background/tissue segmentation model #81

Closed silvandeleemput closed 5 months ago

silvandeleemput commented 6 months ago

This PR adds the WSI background/tissue segmentation model from GC to Mhub. GitHub Repo: https://github.com/DIAGNijmegen/pathology-tissue-background-segmentation-processor GC page: https://grand-challenge.org/algorithms/tissue-background-segmenation/

Algorithm I/O

Input:
- A WSI image (DICOM)
Output:
- A WSI background/tissue segmentation image (TIFF)

Caveats

Since no DicomSeg output is supported for WSI images, the output is currently in TIFF format.
Desired input/output spacings can differ per WSI image, spacings have been made configurable as config options of the runner module.

silvandeleemput commented 6 months ago

gc_wsi_bgseg_output.zip

silvandeleemput commented 6 months ago

/test

sample:
  idc_version: 17.0
  data:
    - SeriesInstanceUID: 1.3.6.1.4.1.5962.99.1.279243577.1019802107.1640956783417.2.0
      aws_url: s3://idc-open-data/64960f35-b3c7-4874-8837-0ec5691e3b63/*
      path: dicom

reference:
  url: https://github.com/MHubAI/models/files/15390723/gc_wsi_bgseg_output.zip

silvandeleemput commented 6 months ago

/review

Review requested. See initial PR comment for caveats.

silvandeleemput commented 6 months ago

gc_wsi_bgseg_output.zip

silvandeleemput commented 6 months ago

/test

sample:
  idc_version: 17.0
  data:
    - SeriesInstanceUID: 1.3.6.1.4.1.5962.99.1.279243577.1019802107.1640956783417.2.0
      aws_url: s3://idc-open-data/64960f35-b3c7-4874-8837-0ec5691e3b63/*
      path: dicom

reference:
  url: https://github.com/MHubAI/models/files/15501333/gc_wsi_bgseg_output.zip

Test Results (24.06.03_10.26.45_9t3n5BDtdf)

```yaml id: ea997558-4f89-4537-84c8-0ba2a6842919 date: '2024-06-03 11:10:22' checked_files: - file: gc_wsi_bgseg.tiff path: /app/test/src/1.3.6.1.4.1.5962.99.1.279243577.1019802107.1640956783417.2.0/gc_wsi_bgseg.tiff checks: - checker: SizeCheck findings: - label: Size Difference description: file size is smaller than reference info: src_size: 349980 ref_size: 349988 diff_size: -8 summary: files_missing: 0 files_extra: 0 checks: SizeCheck: files: 1 findings: Size Difference: 1 conclusion: false ```

silvandeleemput commented 5 months ago

@LennyN95 Regarding the results: description: file size is smaller than reference (349980!=349988) This can surely be the case. small differences are expected due to GPU differences. This is a bit of an edge case due to not being able to convert this to DicomSeg yet for WSI. You can use the ASAP viewer (https://github.com/computationalpathologygroup/ASAP/releases/tag/ASAP-2.2-(Nightly)) to visually inspect and compare the Tiff files.