Workflow Development Guide

Workflow classes coordinate multiple processing steps behind a stable Python API and, where useful, an installed CLI command.

Current Workflow Mapping

CLI command	Workflow class
`monai-physio-convert-image-to-usd`	`monai_physio.WorkflowConvertImageToUSD`
`monai-physio-convert-image-to-vtk`	`monai_physio.WorkflowConvertImageToVTK`
`monai-physio-convert-vtk-to-usd`	`monai_physio.WorkflowConvertVTKToUSD`
`monai-physio-create-statistical-model`	`monai_physio.WorkflowCreateStatisticalModel`
`monai-physio-fit-statistical-model-to-patient`	`monai_physio.WorkflowFitStatisticalModelToPatient`
`monai-physio-reconstruct-highres-4d-ct`	`monai_physio.WorkflowReconstructHighres4DCT`
`monai-physio-train-physicsnemo`	`monai_physio.WorkflowTrainPhysicsNeMo`
`monai-physio-infer-physicsnemo`	`monai_physio.WorkflowInferPhysicsNeMo`
`monai-physio-convert-image-4d-to-3d`	`monai_physio.ConvertImage4DTo3D` (a converter, not a workflow)
`monai-physio-download-data`	`monai_physio.DownloadData` (a utility, not a workflow)
`monai-physio-visualize-pca-modes`	Reads a `pca_model.json` directly; no workflow class

That is all eleven installed commands. Two workflow classes have no CLI wrapper: monai_physio.WorkflowFinetuneICONRegistration and monai_physio.WorkflowEvaluateMovement.

Workflow Example

from pathlib import Path

import itk

from monai_physio import RegisterImagesICON, WorkflowConvertImageToUSD

frame_files = sorted(Path("data/Slicer-Heart-CT").glob("slice_???.mha"))
time_series_images = [itk.imread(str(path)) for path in frame_files]

workflow = WorkflowConvertImageToUSD(
    time_series_images=time_series_images,
    reference_image=time_series_images[0],
    output_directory="./results",
    usd_project_name="patient_001",
    registration_method=RegisterImagesICON(),
)

results = workflow.process()

Adding a Workflow

Inherit from monai_physio.MONAIPhysioBase.
Keep the constructor explicit and typed.
Use self.log_info() and self.log_debug() for runtime status.
Keep file I/O behavior predictable and documented.
Add a CLI wrapper only when the workflow is useful from the command line.
Add focused tests using synthetic data where possible.
Run graphify update . after public API changes - methods added, modified, or removed.

Risk Areas

Changes at the ITK-to-PyVista boundary, time-series transform direction, or LPS-to-USD-Y-up coordinate conversion are high risk and should include focused tests plus visual or metadata validation.

See Also