Prepare the CTA
Normalize orientation and resolution, localize the heart, then crop a 128³ input with the REPRISE-compatible intensity protocol.
Original image stays localORIGINAL V23 · JOINT TPS 600
From CTA.
To a volumetric
heart.
A three-dimensional image becomes a labelled tetrahedral mesh. One surface network. One shared deformation. Ready to explore.
9 predicted surfaces24 template parts1.52M tetrahedra
REPRISE3 / ACQUIRED CARDIAC PHASESV23
20 acquired phases · fixed camera
01 / THE HEART IN MOTION
Geometry that
moves with the heart.
A complete 20-phase sequence, reconstructed with the released inference pipeline. Each phase starts with its own CTA image.
REAL INFERENCE OUTPUT
Follow the full cycle.
We screened three complete sequences and selected one with a clear contraction pattern. The animation shows the 20 exported volumetric meshes in acquisition order.
20 / 20phases reconstructed
0nonpositive tetrahedra in these exports
Independent single-phase V23 inference; no temporal filter or interpolated mesh geometry. Playback speed is illustrative.
02 / IMAGE → SURFACE → VOLUME
A shared field.
A coherent whole.
Predict the surface
The original V23 network and frozen weights predict nine cardiac surfaces, including the Aorta and pulmonary artery.
No default retrainingDeform the volume
Geometric correspondences drive one joint TPS field across all template parts, with Aorta continuation and a tetrahedron determinant guard.
One shared deformation03 / LOOK BENEATH THE SURFACE
Not just a boundary.
An interior.
Move through the cutaway to see the original tetrahedral elements inside the heart walls. The camera orbits and moves closer; the mesh itself is unchanged.
LV wallRV wallLA wallRA wallAortaPA
MMWHS example. The cut removes complete tetrahedra behind a moving plane; no remeshing or display smoothing.
Inspect the scaled Jacobian map +
Fengming Turbo-64 · common range [0, 1]. Element quality alone does not establish anatomical or simulation validity.
04 / BEYOND THE DEVELOPMENT DATA
New CTA.
The same frozen model.
Explore reconstruction examples from MMWHS and ImageCAS with dataset-aware preprocessing and the same released V23 weights.
EXTERNAL CTA / MMWHSOne template,
One template,
new anatomy.
A labelled whole-heart reconstruction viewed from changing angles. No MMWHS fine-tuning was used for this example.
- Surface predictor
- Original V23
- Registration
- Joint TPS 600
- Nonpositive tetrahedra
- 0
EXTERNAL CTA / IMAGECASLocal data.
Local data.
A portable pipeline.
Native-image preprocessing followed by the original V23 model and shared volumetric deformation. No ImageCAS fine-tuning was used for this example.
- Surface predictor
- Original V23
- Registration
- Joint TPS 600
- Nonpositive tetrahedra
- 0
These are selected qualitative development examples. The separate 10 MMWHS + 10 ImageCAS surface evaluation is documented in the repository; it is not a 20-case final-volume validation.
05 / MAKE IT YOURS
Your image.
Your volumetric heart.
Run the code directly, or give your coding agent the HeartVolMesh skill. Installation creates an isolated environment and verifies the model and runtime downloads.
For people
Readable source, preprocessing, mesh export, quality reports and visualization scripts.
Explore the GitHub repository ↗For agents
A GPU-aware bootstrap and an inference workflow for your own local CTA files.
Install heartvolmesh ↗
heartvolmesh / quick start
git clone https://github.com/fmlinks/HeartVolMesh.git
cd HeartVolMesh
python bootstrap.py --workspace ./workspace
python run_inference.py --workspace ./workspace \
--image ./CTA.nii.gz --output ./resultWindows RTX 5090 was tested in a fresh environment. The installer provides a driver-aware RTX 3090 route; that hardware has not been directly tested. Review heart coverage and native CT overlays for each new dataset. This is a research release.