NA-MIC Project Weeks
Back to Projects List
MHub Contributors
Key Investigators
- Leonard Nürnberg (Brigham and Women's Hospital, Boston)
- Andriy Fedorov (Brigham and Women's Hospital, Boston)
- Hugo Aerts (Brigham and Women's Hospital, Boston)
Project Description
MHub.ai is a platform for deep-learning models in medical imaging. We aim to make AI in medical imaging as simple as possible. Therefore, all MHub models need zero set-ups, can be run with a single command, have a standardized IO interface, run directly on DICOM data, are fully customizable to run on other data types and file structures, are tested and reproducible, and run entirely offline. MHub also provides a toolbox to support developers with data conversion, organization, and standardization tasks.
We want to demonstrate WHY bundling models in the MHub standard, make them as simple as possible to use, and provide a valuable resource to the community.
Furthermore, we’re thrilled to show HOW any model or algorithm can be wrapped into an MHub container. We plan to show the process, explain the tools we use, answer questions, and provide assistance and guidance to those who want to use or contribute to an MHub model.
Objective
- Present the MHub.ai platform and model repository with more than 20 models (and counting).
- Demonstrate the benefits of containerized and standardized models and how you can build on them for reproducible research.
- Show how to implement any model in MHub in three steps and provide them to the community.
- Support participants in implementing (their) models into MHub.
- Gather feedback, improve our documentation, and explore what topics, formats, details, and intensity are best for the educational materials.
Approach and Plan
We plan to hold a workshop or break-out session where we demonstrate every step of the contribution process for MHub models in a walk-through style tutorial. We will give detailed examples, discuss best practices, and provide hands-on guidance to all who are planning to implement models into MHub.
Progress and Next Steps
We currently host 27 segmentation, prediction and feature extraction models with 10+ more models under active development.
To help our users contributing models to our platform we provide a detailed documentation and step-by-step tutorials:
-
MHub.ai Documentation We have detailed documentation on how to run a model in MHub and documentation on the individual tools provided within the MHub-IO framework.
- MHub.ai Model Deployment We created a tutorial that guides through the implementation of a model into the universal MHub format.
-
MHub.ai Contribution Process MHub has a clearly defined contribution process. The requirements and the process are explained in our documentation.
- MHub.ai Tutorials We wrote two more tutorials demonstrating how to run and customize MHub models based on public data from IDC and how to visualize and compare results in 3D Slicer.
Illustrations
Background and References
You can learn more about the MHub platform, repository, and framework at the following links.
To dive deeper, you can find the developer documentation, tutorials, and the implementation of all models currently in our repository under these links.