Medical Open Network
for Artificial Intelligence
Maintained by researchers and engineers at NVIDIA, NIH, King's College London, Mayo Clinic, MSKCC, Stanford, DKFZ, and 30+ other institutions.
Why MONAI
Built for medical imaging
Domain-specific transforms, validated 3D architectures, and reproducible workflows. This is not a general ML framework retrofitted for healthcare.
-
PyTorch native
Built on PyTorch, so there is no new framework to learn. Standard
nn.Module,DataLoader, AMP, and DDP all work as you expect. -
Domain-specific tooling
DICOM and NIfTI I/O, spatial transforms for 3D medical volumes, segmentation metrics (Dice, Hausdorff, Surface Distance), and losses calibrated for class-imbalanced anatomy.
-
Reproducible Bundles
MONAI Bundles package weights, training configs, metadata, and inference code together. The same format powers the Model Zoo, so your work is portable from day one.
-
Community-governed
Maintained by NVIDIA, NIH, King's College London, Mayo Clinic, MSKCC, Stanford, DKFZ, and 30+ other institutions. Governance, working groups, and roadmap decisions are public.
-
State-of-the-art architectures
Reference implementations of MAISI (synthetic CT generation), UNETR and SwinUNETR (3D transformer backbones), VISTA-3D (universal segmentation), and Auto3DSeg (automated pipelines).
-
Apache 2.0
Open source under Apache 2.0. Freely usable in commercial products, research, and clinical pipelines, with no hidden licensing gotchas for production use.
Ecosystem
Three projects, one pipeline
Annotate in Label, train in Core, deploy with Deploy. They share one bundle format and the same data conventions, so you can use any piece on its own or all three together.
MONAI Label
Server-side annotation with active learning. It suggests the next volume to label and refines predictions from your corrections. Plugs into 3D Slicer, OHIF, and MITK.
- Active learning for efficient data selection
- Multiple viewer integrations
- AI-assisted annotation
- Multi-user collaboration
MONAI Core
The training and research library. Medical-imaging transforms, 3D architectures, losses, and metrics: the building blocks researchers reach for first.
- Medical-specific transforms
- MAISI, UNETR & VISTA-3D architectures
- Pre-trained model zoo
- Automated ML pipelines
MONAI Deploy
The path from trained model to clinical pipeline. DICOM and FHIR I/O, containerized MAP packaging, and inference runtimes from workstation to Kubernetes.
- Clinical workflow integration
- DICOM & FHIR support
- Containerized deployment
- Inference optimization
Connect
Community & Support
Questions, contributions, and discussions are public. Ask on GitHub or Slack; attend working-group calls; present at the MONAI Bootcamp.
Impact
Success Stories
How hospitals, vendors, and research groups run MONAI in production.
Featured Clinical AI Integration at Mayo Clinic
The Center for Augmented Intelligence at Mayo Clinic Florida runs MONAI-packaged models inside its clinical Radiology pipeline. The case study documents the integration pattern, MAP containerization, and throughput results.
Mayo Clinic Florida · CAII viewer 
Mercure DICOM Orchestration
MONAI Application Packages run inside the Mercure DICOM Orchestrator, so a trained model can join a radiology routing workflow without custom glue code.
Enterprise AI at Scale
Siemens Healthineers adopted MONAI Deploy for their Digital Marketplace, enabling enterprise-scale AI deployment globally.
× Global AI Marketplace
Mayo Clinic AI apps built with MONAI accessible to 10,000+ institutions via Siemens Digital Marketplace.
Contributors
Maintained by the medical-imaging community
These institutions have dedicated engineering, clinical, and research staff to MONAI. Contributions are reviewed and merged in the open.
