Heart Gated CT Processing

Process cardiac gated CT images into dynamic, animated heart models for visualization in NVIDIA Omniverse.

Overview

The physiomotion4d-heart-gated-ct script processes 4D cardiac CT scans through a complete pipeline that includes:

AI-based anatomical segmentation
Deformable image registration across cardiac phases
Surface mesh extraction and transformation
USD file generation with anatomically-realistic materials

This tool is ideal for:

Visualizing cardiac motion through the cardiac cycle
Creating patient-specific animated heart models
Research and educational demonstrations of cardiac physiology
Pre-procedural planning with dynamic anatomical models

Input Requirements

Data Format

Accepts two input formats:

Option 1: Single 4D NRRD file

Contains all cardiac phases in one file
Format: .nrrd or .seq.nrrd
Example: cardiac_4d_ct.nrrd

Option 2: Multiple 3D files (time series)

Separate file for each cardiac phase
Formats: .nrrd, .nii, .mha
Example: phase_00.nrrd, phase_10.nrrd, phase_20.nrrd

Image Quality Requirements

Modality: Cardiac CT (contrast-enhanced recommended)
Phases: Minimum 2, typically 10-20 cardiac phases
Resolution: 0.5-2mm isotropic recommended
Field of View: Must include entire heart and major vessels
Image Quality: Minimal motion artifacts, good contrast

Basic Usage

Single 4D File

physiomotion4d-heart-gated-ct cardiac_4d.nrrd --contrast

Multiple 3D Files

physiomotion4d-heart-gated-ct phase_*.nrrd --contrast --project-name patient_001

With Output Directory

physiomotion4d-heart-gated-ct cardiac.nrrd \
    --contrast \
    --output-dir ./results/patient_001 \
    --project-name patient_001

Command-Line Options

Required Arguments

input_files         Path to 4D NRRD file OR list of 3D files

Optional Arguments

Option	Default	Description
`--output-dir DIR`	`./results`	Output directory for intermediate and final files
`--project-name NAME`	`cardiac_model`	Project name used for USD file naming
`--contrast`	False	Flag for contrast-enhanced studies (recommended)
`--reference-image PATH`	Auto (70%)	Custom reference image for registration
`--registration-iterations N`	1	Number of registration refinement iterations
`--registration-method`	`icon`	Registration method: `icon` or `ants`

Processing Pipeline

The script executes these steps automatically:

Load Time Series * Converts 4D to 3D time series OR loads multiple files * Selects reference image (default: 70% cardiac phase)
Segmentation * Segments reference image using AI segmentation (TotalSegmentator or Simpleware) * Identifies: heart chambers, myocardium, vessels, lungs, bones
Registration * Registers each phase to reference image * Creates separate registrations for dynamic vs static anatomy * Uses mass-preserving deformable registration
Mesh Generation * Extracts VTK surface meshes from segmentation * Creates smooth anatomical surface representations
Transform Application * Applies registration transforms to meshes at each time point * Generates animated mesh sequences
USD Creation * Produces painted USD files with anatomical materials * Creates time-varying geometry for Omniverse

Output Files

Primary Outputs

Located in <output-dir>/:

<project_name>.dynamic_anatomy_painted.usd
    Animated heart and vessels with cardiac motion

<project_name>.static_anatomy_painted.usd
    Stationary anatomy (lungs, bones, tissues)

<project_name>.all_anatomy_painted.usd
    Combined model with all structures

Intermediate Files

intermediate/
├── slice_*.mha                  # Individual 3D images per time point
├── slice_*.labelmap.mha         # Segmentation masks
├── slice_*.reg_*.inverse_transform.hdf5  # Registration transforms
└── slice_max.reg_*.mha          # Maximum intensity projections

meshes/
├── *.vtk                        # VTK mesh files
└── *_4d.vtk                     # Time series VTK files

Examples

Contrast-Enhanced Cardiac CTA

physiomotion4d-heart-gated-ct cardiac_cta.nrrd \
    --contrast \
    --output-dir ./output/patient_123 \
    --project-name PatientXYZ_CTA

Non-Contrast Study with Custom Reference

physiomotion4d-heart-gated-ct phase_*.nrrd \
    --reference-image phase_05.mha \
    --output-dir ./results \
    --project-name noncontrast_cardiac

Research Dataset Processing

# Batch process multiple cases
for case_dir in /data/cardiac_studies/case_*/; do
    case_name=$(basename "$case_dir")
    physiomotion4d-heart-gated-ct ${case_dir}/*.nrrd \
        --contrast \
        --output-dir ./results/${case_name} \
        --project-name ${case_name}
done

With ANTs Registration

physiomotion4d-heart-gated-ct cardiac.nrrd \
    --contrast \
    --registration-method ants \
    --registration-iterations 50

Best Practices

Choosing Reference Image

Default (70% phase): Mid-diastole, suitable for most cases
Custom selection: Choose phase with:
- Best image quality
- Least motion artifact
- Optimal vessel enhancement (for CTA)

Performance Optimization

GPU: Enable CUDA for faster registration (automatic if available)
RAM: 16GB+ recommended for large datasets
Storage: Use SSD for faster I/O operations
Iterations: Start with default (1), increase only if needed

Data Quality Tips

Verify cardiac phases are temporally ordered
Check consistent spacing and orientation across phases
Ensure uniform contrast enhancement (for CTA)
Confirm heart structures are within field of view

Troubleshooting

Input File Issues

Error: “Input file not found”

Verify file paths are correct
Use absolute paths if relative paths fail
Check file permissions

Error: “Unable to read image”

Confirm file format is supported (NRRD, NII, MHA)
Check file is not corrupted
Verify image metadata is complete

Segmentation Quality Issues

Poor heart segmentation

Use --contrast flag for CTA studies
Verify cardiac structures are visible
Check image quality (no severe artifacts)

Missing anatomical structures

Ensure structures are in field of view
Check image resolution is adequate
Verify reference image quality

Registration Failures

Registration not converging

Try different reference image (better quality phase)
Increase --registration-iterations
Switch registration method (--registration-method ants)

Excessive deformation

Verify sufficient temporal overlap between phases
Check image spacing is reasonable
Ensure phases are properly ordered

Memory Issues

Out of memory errors

Reduce image resolution if possible
Process fewer time points
Close other applications
Consider using machine with more RAM

See Troubleshooting for additional help.

Expected Processing Time

Approximate times on typical workstation (32GB RAM, NVIDIA GPU):

Dataset	GPU Available	No GPU
10 phases, 512³ volume	15-30 minutes	2-4 hours
20 phases, 512³ volume	30-60 minutes	4-8 hours

Viewing Results in Omniverse

Open NVIDIA Omniverse
Navigate to output directory
Open .usd file
Press Play to view cardiac motion animation
Adjust timeline speed for desired visualization

Next Steps

See Best Practices for optimization strategies
Review Troubleshooting for common issues
For Python API access, see physiomotion4d.WorkflowConvertHeartGatedCTToUSD in Workflow Development Guide