NA-MIC Project Weeks

Back to Projects List

OMAS CT: Open Model for Anatomy Segmentation in Computer Tomography

Key Investigators

• Tamaz Amiranashvili (University of Zurich, Switzerland)
• Murong Xu (University of Zurich, Switzerland)
• Bjoern Menze (University of Zurich, Switzerland)

Presenter location: In-person

Project Description

We have developed a state-of-the-art automated segmentation model capable of identifying ~170 anatomical structures in volumetric CT scans. This model has been trained on a combined dataset of more than 22,000 diverse, partially-annotated CT scans, setting a new benchmark in medical imaging. Our goal is to integrate this model into a 3D Slicer extension, making it widely available to the community.

Objective

1. Model development: Train models on partially-annotated datasets for whole-body CT segmentation covering approximately 170 structures.
2. Open source the trained models: Open-source the trained models and the associated codebase on 3D Slicer and other platforms, making them easily accessible and utilizable for clinical and research purposes, among others.
3. Release the data: Release the expansive dataset and corresponding annotations on the Imaging Data Commons (IDC), facilitating further research on medical image analysis.

Approach and Plan

1. Data Management: Collection and curation of CT scans.
2. Model Training and Evaluation: Systematic training and assessment of models.
3. Data Release: Consolidation and release of the dataset and corresponding annotations in appropriate formats (e.g., DICOM) on IDC.
4. Model Release: Publication of final model weights.
5. Software Integration: Development and integration of a module for 3D Slicer, optimized for both CPU and GPU usage to accommodate varying user hardware.
6. Documentation: Creation of detailed user guidelines to facilitate the easy application of the models.

Progress and Next Steps

Current Achievements:

1. Prototypes of the trained models and an operational inference pipeline have already been developed.

In progress / next steps:

1. Benchmarking on public medical image segmentation challenges, followed by evaluation and analysis of results.
2. Preparing the dataset and labels for public release.
3. Developing the 3D Slicer plugin for integration.

Illustrations

Background and References

TBD