MONAI
An open-source medical imaging AI framework built on PyTorch, with 3D image processing, pre-trained models, and training across multiple GPUs.
MONAI is a framework for researchers and developers building AI models for medical images. Its Core library provides medical-specific tools for training and evaluating models within existing PyTorch workflows.
- Medical image processing: Read DICOM and NIfTI data and apply spatial transforms to 3D medical volumes, with segmentation metrics and losses suited to imbalanced anatomical classes.
- Models and automated pipelines: Use pre-trained models from the Model Zoo, Auto3DSeg for automated segmentation pipelines, and architectures such as UNETR, SwinUNETR, VISTA-3D and MAISI for segmentation or synthetic CT generation.
- Portable model bundles: MONAI Bundles keep model weights, training settings, metadata and inference code together so researchers can share and reproduce workflows.
- Distributed training: Train across multiple GPUs and machines while retaining PyTorch's standard model and data-loading interfaces.
MONAI is open source under Apache 2.0. Core is a Python library that depends on NumPy and PyTorch, and Docker images support containerized use on your own infrastructure.
Annotation and clinical deployment belong to separate projects in the ecosystem. MONAI Label provides server-side, AI-assisted annotation and active learning through 3D Slicer, OHIF and MITK. MONAI Deploy packages inference applications for workstations or Kubernetes, with DICOM and FHIR support. You can use these projects independently or combine them with Core.
