HI-ML open-source toolbox
Open-source tools to help simplify deep learning models for healthcare and life sciences
The Health Intelligence Machine Learning (HI-ML) OSS toolbox helps to simplify and streamline work on deep learning models for healthcare and life sciences by providing tested components (data loaders, pre-processing), deep learning models, and cloud integration tools. It is created and used for machine learning (ML) research by multiple groups (opens in new tab) in Microsoft Health Futures (opens in new tab). It is released at no-cost under an MIT open-source license and supports the FAIR Principles for open science (opens in new tab) to make it widely available for the global healthcare machine learning community, who can leverage our work.
Open source
HI-ML is open source, based on PyTorch, and released under an MIT license.
Easy to use
Makes building healthcare ML models easier, increasing productivity of scientists, clinicians, and engineers.
Scalable
Uses Microsoft Azure to train models at scale using the latest GPU technology.
Research applications
HI-ML can be used to make ML model development easier for a variety different health and life sciences research applications, including:
Multi-modal radiology
HI-ML makes it easy to work with multimodal image & text data, such as chest x-ray images and associated radiological reports. We have released pre-trained ML models on Hugging Face and associated datasets on PhysioNet as part of our ECCV 2022 paper “Making the Most of Text Semantics to Improve Biomedical Vision–Language Processing“
Histopathology
HI-ML provides dedicated tools for end-to-end histopathology ML model development, testing, and visualization in Microsoft Azure.
Open-source toolkits and components
The HI-ML OSS toolbox comprises several packages and components to increase the productivity of health and life science researchers and developers. You can find out more on Read the Docs (opens in new tab). If you have any problems, find issues in the code, or have a feature request, then please create an issue on GitHub (opens in new tab). We monitor these issues and will look to respond via GitHub.
HI-ML (opens in new tab)
Python package providing ML components for healthcare machine learning.
HI-ML-Azure (opens in new tab)
Python package providing helper functions for running in Azure Machine Learning.
HI-ML-cpath (opens in new tab)
Python package and ML model training workflows for working with histopathology images.
HI-ML-multimodal
Python package for working with multi-modal health data.
Who can benefit from the HI-ML OSS toolbox?
- Academic and clinical researchers, including at Academic Medical Centers, can focus on their research by using the HI-ML OSS toolbox to be more productive.
- Medical imaging companies, who can use the HI-ML OSS toolbox to help to accelerate development of healthcare-related ML models at scale using Microsoft Azure.
Related projects
- Project InnerEye open-source software (OSS) is created and used for deep learning research by the Project InnerEye team at Microsoft Research (MSR) Cambridge, UK. It is released at no-cost under an MIT open-source license to make it widely available for the global medical imaging community, who can leverage our work. The tools aim to increase productivity for research and development of best-in-class 3D medical imaging segmentation AI and includes tools to test these models in clinical settings (subject to appropriate regulatory approvals).
- HI-ML makes extensive use of Azure Machine Learning to increase productivity by enabling training on GPU clusters, MLOps, and image labelling. For more details, see Azure Machine Learning – ML as a service | Microsoft Azure (opens in new tab)
- The HI-ML OSS toolbox is an open-source research project and not a Microsoft product. It takes advantage of Microsoft Azure to make it easier to develop machine learning models. For more information about Microsoft products, see Azure for Healthcare—Healthcare Solutions | Microsoft Azure (opens in new tab)