Written and maintained by CASRAI Editorial Board
Last updated
MONAI (Medical Open Network for AI) is an open-source, PyTorch-based framework for deep learning in healthcare imaging, released under the Apache-2.0 licence. It is a research software library, not a clinical product and not medical advice. This page is for researchers and lab teams deciding whether to build imaging models on it, and covers what the project contains, who stands behind it, how to install it reproducibly and how to cite it.
What is MONAI (Medical Open Network for AI)?
The project’s GitHub repository describes MONAI as “a PyTorch-based, open-source framework for deep learning in healthcare imaging” and as part of the PyTorch Ecosystem. Its stated features are flexible pre-processing for multi-dimensional medical imaging data, composable and portable APIs, domain-specific networks, losses and metrics, customisation for different levels of expertise, and multi-GPU and multi-node data parallelism (GitHub README, Project-MONAI/MONAI, read 2026-11-07).
The MONAI paper (Cardoso et al., 2022, arXiv:2211.02701) puts it in one line: MONAI extends PyTorch to support medical data, with a particular focus on imaging. It describes a community-supported, consortium-led project that provides medical-specific model architectures, transforms and utilities and follows software best practice. For background on the image types involved, see what medical imaging is.
Meaning of the name
MONAI stands for Medical Open Network for AI. The Project-MONAI GitHub organisation uses the name for the whole project, while the library called “MONAI” or “MONAI Core” is one repository inside it.
How is the name pronounced?
The project’s README and paper give the name only as an acronym of Medical Open Network for AI and, in the sources checked, do not state a pronunciation, so this page does not give one.
Who develops MONAI and what does the project cover?
MONAI was announced by NVIDIA and King’s College London in 2020, according to the title of NVIDIA’s announcement post and contemporaneous trade coverage, which describe it as built on the Ignite and PyTorch frameworks and drawing on earlier tools including NVIDIA Clara, NiftyNet, DLTK and DeepNeuro. Today the code lives in the Project-MONAI GitHub organisation, and the README states the goal of a “community of academic, industrial and clinical researchers collaborating on a common foundation.” The 2022 paper lists 57 authors.
The organisation hosts several separate pieces, each described in its own repository:
| Component | Description as stated by the project | Who it mainly serves |
|---|---|---|
| MONAI (MONAI Core) | “AI Toolkit for Healthcare Imaging”; the PyTorch-based library described above | Researchers training and evaluating imaging models in Python |
| MONAI Label | “An intelligent open source image labeling and learning tool” | Teams that need annotated imaging data |
| MONAI Deploy App SDK | “A framework and associated tools to design, develop and verify AI-driven applications in the healthcare imaging domain” | Teams packaging imaging AI as applications |
| MONAI Tutorials | Listed as “MONAI Tutorials” | New users learning the API |
| MONAI Model Zoo | Hosts models in the MONAI Bundle format | Users looking for pre-built models |
Source: github.com/Project-MONAI, read 2026-11-07. The third column is an editorial reading of those descriptions.
How do you install and start using MONAI?
The README gives three installation routes. These steps come from the project README and PyPI listing as read on 2026-11-07.
- Check your Python version. The PyPI listing declares Python 3.10 or later.
- Install the released package from PyPI with
pip install monai. The README notes that the development branch is unreleased and that releases are published on PyPI. - Alternatively, install from conda-forge, which the README lists as a supported route.
- Alternatively, run the project’s Docker image:
docker run -ti --rm --gpus all projectmonai/monai:latest /bin/bash. - Work through the Project-MONAI/tutorials repository to see transforms, networks, losses and metrics used together.
- Consult the API documentation for the exact signatures of the version you installed.
Using MONAI reproducibly
For a published analysis, record the exact released version (pip show monai prints it) and prefer a pinned Docker image tag over latest, which moves. The general practice is covered in reproducibility infrastructure: workflows, containers and code sharing; if your cluster uses Apptainer rather than Docker, see Apptainer vs Docker. This is editorial advice built on the README’s statement that releases are versioned on PyPI and that a Docker image is provided.
Strengths and limits
- Open licence. Apache-2.0 (README and PyPI metadata, read 2026-11-07), so it can be used and modified in academic and industrial work subject to that licence’s terms. See the open-source software definition for the general concept.
- Medical-specific building blocks on PyTorch. Transforms, networks, losses and metrics for medical imaging data, rather than general computer-vision defaults.
- Scales across GPUs and nodes. Multi-GPU and multi-node data parallelism is a listed feature.
- Requires Python and PyTorch skills. It is a code library described as an extension of PyTorch, so users work in Python.
- A research framework, not a regulatory clearance. The sources read describe a software toolkit for research and application development. Whether a model built with it counts as a regulated device depends on its intended use; see software as a medical device (SaMD).
MONAI compared with using plain PyTorch
| Question | Plain PyTorch | MONAI |
|---|---|---|
| Relationship | The underlying deep-learning library | Extends PyTorch for medical data, with a focus on imaging (Cardoso et al., 2022) |
| Medical-specific transforms, networks, losses, metrics | You write or source them yourself | Provided as domain-specific components (README) |
| Annotation and deployment tooling | Outside the library | Separate MONAI Label and MONAI Deploy App SDK projects |
How do you cite MONAI?
Cite the project paper: Cardoso, M. J., et al. (2022). MONAI: An open-source framework for deep learning in healthcare. arXiv:2211.02701, DOI 10.48550/arXiv.2211.02701 (read 2026-11-07). Also state the software version you ran. General rules for software citation are in how to cite software, code and R packages, and if you report a medical AI study, the MINIMAR minimum reporting checklist for medical AI covers what else to disclose.
Frequently asked questions
What is Medical Open Network for AI (MONAI)?
An open-source, PyTorch-based framework for deep learning in healthcare imaging, hosted in the Project-MONAI GitHub organisation and part of the PyTorch Ecosystem.
Who developed MONAI?
NVIDIA and King’s College London announced it in 2020. It is now described as a community-supported, consortium-led project with academic, industrial and clinical contributors (Cardoso et al., 2022).
Is MONAI free to use?
Yes. The code is licensed under Apache-2.0 and installs from PyPI with pip install monai. The licence has its own conditions on redistribution and notices, so read it before redistributing modified code. Hardware such as GPUs is your own cost.
What is the difference between MONAI, MONAI Label and MONAI Deploy?
MONAI (Core) is the training and evaluation library; MONAI Label is an image labelling and learning tool; MONAI Deploy App SDK is for building and verifying imaging AI applications. See the component table above.
How do you pronounce MONAI?
No primary source checked states an official pronunciation. MONAI is an acronym for Medical Open Network for AI, and the project documentation gives no spoken form.
Sources
- Project-MONAI/MONAI README, github.com/Project-MONAI/MONAI (read 2026-11-07)
- Project-MONAI organisation page, github.com/Project-MONAI (read 2026-11-07)
- Cardoso, M. J. et al. (2022), MONAI: An open-source framework for deep learning in healthcare, arxiv.org/abs/2211.02701 (read 2026-11-07)
- PyPI listing for monai, pypi.org/project/monai (read 2026-11-07)
- NVIDIA and King’s College London Announce MONAI Open Source AI Framework for Healthcare Research, blogs.nvidia.com (title and trade-press coverage, checked 2026-11-07)
