[Paper Review] MONAI: An open-source framework for deep learning in healthcare

M. Jorge Cardoso, Wenqi Li+55 more|arXiv (Cornell University)|Nov 4, 2022

Artificial Intelligence in Healthcare and Education426 citations

TL;DR

MONAI Core is a PyTorch-based, open-source framework that provides domain-optimized components, transforms, networks, and workflows for deep learning in healthcare, with a strong community and consortium backing.

ABSTRACT

Artificial Intelligence (AI) is having a tremendous impact across most areas of science. Applications of AI in healthcare have the potential to improve our ability to detect, diagnose, prognose, and intervene on human disease. For AI models to be used clinically, they need to be made safe, reproducible and robust, and the underlying software framework must be aware of the particularities (e.g. geometry, physiology, physics) of medical data being processed. This work introduces MONAI, a freely available, community-supported, and consortium-led PyTorch-based framework for deep learning in healthcare. MONAI extends PyTorch to support medical data, with a particular focus on imaging, and provide purpose-specific AI model architectures, transformations and utilities that streamline the development and deployment of medical AI models. MONAI follows best practices for software-development, providing an easy-to-use, robust, well-documented, and well-tested software framework. MONAI preserves the simple, additive, and compositional approach of its underlying PyTorch libraries. MONAI is being used by and receiving contributions from research, clinical and industrial teams from around the world, who are pursuing applications spanning nearly every aspect of healthcare.

Motivation & Objective

• Clarify the design goals and challenges of deep learning in healthcare and the need for a standardized, robust software framework.
• Describe MONAI Core’s architecture, design principles, and how it integrates with the PyTorch ecosystem.
• Explain the domain-specific components (transforms, datasets, networks, engines) that support medical imaging pipelines.
• Highlight the open-source strategy, governance, and ecosystem (MONAI Label/Deploy/Consortium) that enable collaboration and deployment.
• Illustrate use cases and applications across healthcare AI research and potential clinical deployment.

Proposed method

• Present MONAI Core as a PyTorch-based, opt-in framework that integrates with the PyTorch ecosystem and Ignite for training loops.
• Describe core modules (data, losses, networks, transforms, visualization, metrics, optimizers) and workflow engines/handlers for training and inference.
• Detail domain-specific transforms and physics-informed, invertible, and dictionary-based data operations for medical imaging.
• Explain extended dataset handling (MetaTensor, CacheDataset, PersistentDataset) and dataset wrappers for standard medical benchmarks.
• Outline the open-source strategy (Apache-2.0), community collaboration, and surrounding MONAI components (Label, Deploy, Federated Learning, Education).
• Discuss training/inference workflows, distributed data-parallel support, and compatibility with PyTorch distributed, Horovod, XLA, and SLURM.

Experimental results

Research questions

• RQ1 How does MONAI Core address the unique data, metadata, and workflow requirements of medical imaging in deep learning?
• RQ2 In what ways does MONAI integrate with and extend the PyTorch ecosystem to enable reproducible, scalable healthcare AI development?
• RQ3 What domain-specific components (transforms, datasets, networks, engines) does MONAI provide to streamline medical AI research and deployment?
• RQ4 How does MONAI’s open-source strategy and consortium governance facilitate collaboration and translation of research into clinical practice?

Key findings

• MONAI Core provides modular, PyTorch-like components across data, losses, networks, transforms, engines, and metrics to support healthcare AI pipelines.
• It introduces domain-specific transforms (including physics-based and invertible transforms) and MetaTensor to store medical metadata and transformation history.
• The framework offers extended dataset utilities (CacheDataset, PersistentDataset) and dataset wrappers for MedNIST and Medical Segmentation Decathlon, aiding rapid training and benchmarking.
• MONAI training/inference workflows extend Ignite engines with specialized handling and default training utilities, supporting distributed data-parallel and various backend integrations.
• MONAI is open-source under Apache-2.0, fosters a large community of thousands of users and hundreds of contributors, and includes related projects (MONAI Label, Deploy, Federated Learning, Education).
• The design emphasizes compatibility with PyTorch, opt-in incremental adoption, and seamless collaboration with other healthcare tools, accelerating research and potential clinical deployment.