NA-MIC Project Weeks

Back to Projects List

MHub Tutorials and Structured Report Support

Key Investigators

• Leonard Nürnberg (Maastricht University)
• Andrey Fedorov (BWH, USA)
• David Clunie (Pixelmed Publishing, USA)

Presenter location: Onsite

Project Description

Objective

1. Objective A. Implement a new MHub core module to export non-segmentation data as DICOM Structured Report.
2. Objective B. Offer support for interested developers explaining how to contribute models to MHub and how to use and configure MHub models for easy and reproducible research.

Approach and Plan

Objective A

1. Examine the requirements and limitations of DICOM Structured Reports.
2. Examine existing libraries and frameworks that can help in the creation of DICOM Structured Reports.
3. Check to what extent the internal data representation of MHub complies with the requirements of DICOM Structured Reports and, if necessary, plan an extension of the internal data representation.

Objective B

1. Organize a break-out session for all interested developers. We can provide tutorials and answer specific questions on request. It would be great to know who is interested in submitting a model to MHub or using MHub models and gather questions in advance to provide customized tutorials.

Other items suggested by @fedorov:

• revisit MHub Colab notebook
• continue discussions on how the specific series from IDC can be linked with the models, and how this can be exposed in the website and via some API

Progress and Next Steps

1. Developers of MHub models must wrap all non-file outputs (e.g. prediction results or classification labels) in mhub-io objects and describe their semantics.
2. MHub currently offers a JSONExporter module that can be used to query MHub’s internal data representation and generate a customizable json (or csv).

Results from PW40

1. MHub SR Support
   1. We will start by adding support the TID1500 DICOM SR template.
   2. We will use parametrization of highdicom codes for simple annotation of model data.
2. We developed new MHub tutorials for
   1. T1 - Run TotalSegmentator on IDC Collection
   2. T2 - Run Custom MHub Lung Segmentation Workflow on Chest CT in Nifti Format
3. We now have a full documentation for contributions to MHUb (use-case tutorial will follow soon ;))
   1. How to Contribute a Model to MHub

Illustrations

Background and References

Documentation

• MHub documentation
• MHub Handling Logical Output
• MHub Report Exporter Documentation

Code

• MHub-IO Logical Output

Other stuff

• JSON Representation of DICOM Structured Reports (DICOM Supplement 219 Trial Use Draft) including slides explaining the purpose and sample Java implementation of DICOM SR to/from JSON in PixelMed toolkit.
• dcmqi can be used to create TID 1500 structured reports containing measurements derived from segmentations; this is the tool that was used in the recent paper/code for saving radiomics features: https://github.com/ImagingDataCommons/nnU-Net-BPR-annotations