PhysicsNeMo AI Surrogates

PhysioTwin4D trains and runs PhysicsNeMo mesh-stage models: given a subject’s shape parameters and a stage (a point in the cardiac or respiratory cycle), predict a per-vertex target on the shared template mesh. When that target is a displacement, the prediction replaces a per-phase registration solve with one forward pass — see Tutorials 9 through 13 in Tutorials.

The layer follows the same has-a shape as the rest of the workflow tier: a workflow owns the data and the artifacts, and a method object owns the network.

Class	Role
WorkflowTrainPhysicsNeMo	Manifests, normalization, lazy datasets, checkpoints and metadata
WorkflowInferPhysicsNeMo	Loads a trained model and predicts raw per-point targets
WorkflowInferMovement	Interprets 3-component targets as displacements: deformed meshes, mm error statistics, rasterized deformation fields, warped images and USD
WorkflowEvaluateMovement	Scores those predictions per structure against the acquired frames: volume difference, Dice and surface RMSE
TrainPhysicsNeMoMGN / TrainPhysicsNeMoMLP	The networks to train: MeshGraphNet or fully connected
InferPhysicsNeMoMGN / InferPhysicsNeMoMLP	The matching networks at inference time

PhysicsNeMo is an optional dependency:

pip install "physiotwin4d[physicsnemo]"
pip install torch-geometric          # MeshGraphNet only

It requires Python >= 3.11. import physiotwin4d works without it; the imports happen lazily inside the methods that need them.

• The Per-Subject Manifest
  • Targets are read verbatim
  • Reference
  • Supporting helpers
  • See Also
• Training a Mesh-Stage Model
  • Workflow
  • Training methods
  • Example
  • Notes
  • See Also
• Predicting With a Mesh-Stage Model
  • Generic prediction
  • Displacement interpretation
  • Inference methods
  • Example
  • Notes
  • See Also
• Scoring a Mesh-Stage Model Against Images
  • Per-structure scoring
  • Example
  • Notes
  • See Also