Model Registration Developer Guide

Model registration aligns template meshes to patient surfaces and masks. The supported high-level entry point is monai_physio.WorkflowFitStatisticalModelToPatient.

Recommended Entry Point

import itk
import pyvista as pv

from monai_physio import WorkflowFitStatisticalModelToPatient

workflow = WorkflowFitStatisticalModelToPatient(
    template_model=pv.read("template_heart.vtu"),
    patient_models=[pv.read("lv.vtp"), pv.read("rv.vtp")],
    patient_image=itk.imread("patient_ct.nii.gz"),
)

result = workflow.process()

Lower-Level Classes

The workflow composes these lower-level registration classes:

• monai_physio.RegisterModelsICP

• monai_physio.RegisterModelsICPITK

• monai_physio.RegisterModelsDistanceMaps

• monai_physio.RegisterModelsPCA

Use these directly only when developing or testing a specific registration stage. Their constructors and return dictionaries are documented in Model Registration Modules.

Development Notes

• Prefer PyVista mesh objects at the public Python boundary.

• Convert volumetric meshes to surfaces before surface registration when needed.

• Treat ITK/PyVista coordinate transforms as high-risk and add focused tests.

• Keep synthetic test meshes small and deterministic.

• RegisterModelsPCA returns moving_to_fixed_transform / fixed_to_moving_transform. These are point transforms; see Transform Direction Conventions before applying them to images or meshes.

See Also

• Transform Direction Conventions

• Model Registration Modules

• Workflow Development Guide