NA-MIC Project Weeks
Back to Projects List
Pathology Extension for MHub IO Modules
Key Investigators
- Curtis Lisle (KnowledgeVis, LLC, USA)
- Leonard Nürnberg (MGB, Harvard, The Netherlands)
Project Description
Pathology (DICOM) images differ greatly from radiology images, e.g., contain multiple resolutions. The MHub core provides 18 IO Modules to import, convert, organize, imaging data. We want to extend this with additional IO Modules to extract a target resolution and to provide an alternative toolchain to generate DICOMSEG output files. The IO Modules will be made publicly available as an MHub.ai module extension.
Objective
- A public pathology extension for MHub.ai
- Adding the RMS model to MHub.ai utilizing the provided IO Modules as a PoC
Approach and Plan
- Create a new pathology extension repository
- Implement an extractor module
- Implement a specific dicomseg conversion module (e.g., based on highdicom)
- Implement the RMS model as PoC
Progress and Next Steps
- A PathologyExtension repository was created. Modules defined in the extension are automatically discovered during the MHubIO run setup so these extensions are available to all MHub models
- A PathologyResolutionFilter module has been developed and tested on DICOM-WSI images from the NCI Imaging Data Commons. The module reads image metadata and copies only image resolutions that match an input parameter of desired resolutions. Only image resolutions matching the desired target resolution are copied from the input for MHub pipeline processing.
- A prototype method using the HighDicom Python library was developed and tested to write DICOM DSeg (segmentation) images. Next, this code needs to be wrapped as an MHub DSegWriter method and added to the PathologyExtension repository. Once this is complete, pathology algorithns that don’t know how to read and write WSI-DICOM can be used in DICOM to DICOM workflows