How project-monai/monai works
MONAI is a PyTorch-based framework for medical-imaging AI. A typical user workflow loads imaging data, composes preprocessing transforms, trains or evaluates a network, and applies inference and metrics; MONAI also provides reusable losses, visualization, and specialized application workflows. The graph emphasizes those product capabilities and the metadata-aware data layer. Connections are limited to relationships documented in the README or visible in the supplied source index; unsampled application links are omitted.
Main entry points: compose.py Transform composition · trainer.py Training engine · dataset.py Datasets and readers · meta_tensor.py Metadata tensors
Components 22
Imaging data
- Datasets and readers
monai/data/dataset.py - Metadata tensors
monai/data/meta_tensor.py - Box utilities
monai/data/box_utils.py - Shared type definitions
monai/config/type_definitions.py - Runtime utilities
monai/utils/misc.py
Transforms
- Transform composition
monai/transforms/compose.py - Spatial transforms
monai/transforms/spatial/array.py - Intensity transforms
monai/transforms/intensity/array.py - Transform interfaces
monai/transforms/transform.py
Model development
- Neural networks
monai/networks/nets - Network components
monai/networks/blocks - Training losses
monai/losses/dice.py
Training and inference
- Training engine
monai/engines/trainer.py - Inference strategies
monai/inferers/inferer.py - Evaluation metrics
monai/metrics - Visualization
monai/visualize
Specialized applications
- Portable bundles
monai/bundle/config_parser.py - Auto3DSeg
monai/auto3dseg/analyzer.py - Federated learning
monai/fl/client/client_algo.py - Pathology tools
monai/apps/pathology - MRI reconstruction
monai/apps/reconstruction
Other components
- Researcher
Connections 27
- Researcher → Datasets and readers: loads data
- Datasets and readers → Metadata tensors: uses metadata
monai/data/dataset.py - Datasets and readers → Transform composition: uses transforms
monai/data/dataset.py - Researcher → Transform composition: preprocesses images
- Transform composition → Spatial transforms: composes
monai/transforms/__init__.py - Transform composition → Intensity transforms: composes
monai/transforms/__init__.py - Spatial transforms → Transform interfaces: implements
monai/transforms/spatial/array.py - Intensity transforms → Transform interfaces: implements
monai/transforms/intensity/array.py - Spatial transforms → Metadata tensors: updates metadata
monai/transforms/spatial/array.py - Researcher → Neural networks: builds models
- Neural networks → Network components: uses components
- Researcher → Training engine: trains models
- Training engine → Datasets and readers: uses data
monai/engines/trainer.py - Training engine → Transform composition: uses transforms
monai/engines/trainer.py - Training engine → Inference strategies: uses inferer
monai/engines/trainer.py - Researcher → Training losses: chooses objectives
- Researcher → Evaluation metrics: evaluates results
- Researcher → Inference strategies: runs inference
- Inference strategies → Neural networks: uses networks
monai/inferers/inferer.py - Researcher → Visualization: inspects outputs
- Researcher → Portable bundles: configures workflows
- Researcher → Auto3DSeg: runs Auto3DSeg
- Researcher → Federated learning: runs distributed learning
- Researcher → Pathology tools: uses pathology tools
- Researcher → MRI reconstruction: uses reconstruction
- Metadata tensors → Shared type definitions: uses types
monai/data/meta_tensor.py - Metadata tensors → Runtime utilities: uses utilities
monai/data/meta_tensor.py