NA-MIC Project Weeks

Back to Projects List

wsidicomizer-based conversion into DICOM WSI format for Imaging Data Commons

Key Investigators

• Andrey Fedorov (BWH, USA)
• David Clunie (PixelMed Publishing, USA)
• Daniela Schacherer (Fraunhofer MEVIS, Germany)

Project Description

This project is motivated by the operational needs of the Imaging Data Commons to have a sustainable mechanisms supported by open-source software for harmonizing various slide microscopy images in vendor-specific representations into DICOM Whole Slide Imaging format. Our current procedures is not sustainable, since they rely on the PixelMed tools by David Clunie, which are based on a coding style and build process that has not changed since initiated over 20 years ago, unfortunately have “bus factor” of 1, and are not open for community contributions. We started using wsidicomizer as a replacement, but there are gaps in the functionality to be remedied, and additional testing that needs to be done before we can consider operational switch.

This project is the continuation of the earlier project at PW42 2025.

Objective

1. Implement routing of the DICOM metadata.
2. Evaluate extensibility of the conversion approach to various input types.

Approach and Plan

1. Improve the conversion script developed for the HMS SARDANA sample to handle metadata.
2. Evaluate conversion of the .czi samples.
3. Evaluate metadata routing in the Mirax conversion script

Progress and Next Steps

CSV Files → CCDIMetadataLoader → DomainMetadata → MetadataBuilder → WsiDicomizerMetadata + Dataset
                  ↓                                       ↓
            UIDRegistry                           Code Tables (CSV)
             (SQLite)

1. Set up exploration repository that included code base from various related conversion tools and conversion scripts I worked on earlier
2. Identified relevant sample earlier converted in IDC from the CCDI-MCI collection, which was selected as the initial target for development
3. Used Claude Code and Copilot to independently develop plan and initial implementation for migrating from pixelmed (see copilot_solution and claude_solution) in the repo
4. Used Claude Code and Copilot to independently scritinize both solution and summarize pros and cos (see reports in the top level of the repo)
5. Examined the analysis, implementations, selected Copilot solution as the preferred (supported by the analysis)
6. Iterated to make the converter work on the selected CCDI-MCI sample
7. Iterated to add features and refine organization
8. Confirmed functionality on an independent sample not supplied to the agent during development
9. Confirmed dciodvfy DICOM validator does not report issues related to the specimen metadata.
10. Confirmed converted images load with the QuPath BioFormats DICOM loader
11. Next steps:
    • validate specimen metadata against the pixelmed-conversion results
    • look into ICC profile handling, pixel data total size difference, potentially missing label/overview images
    • extend/test for collections other than CCDI

Illustrations

No response

Curious examples of failures of Claude

• remind to check documentation in the prompt?

• need to understand the code …

Background and References

No response