THORAX + CIRCULATION · 8 SEPTEMBER 2026
The heart has its surroundings.
The new organs now contribute to ultrasound. An animated pericardium surrounds the heart, and a faint thorax overlay is optional. Every supplied mesh is usable for this prototype.
New: individual vessel openings and particle transport corrections · view five-angle captures ↗
Reference circulation corrected; moving-valve flow remains approximate. The shared reference interfaces carry the requested flow and both sides pump equal cycle totals. The animated valve apertures still do not support quantitative clinical flow claims.
Watch the integrated scene
Two source cycles, captured offline from the shipped WebGPU shaders and ECG renderer. The ghost overlay is enabled for this recording; it is off by default in the simulator. This is not a browser performance measurement.
Surrounding tissue changes the scan
Actual native GPU output, at 3 MHz and high sampling. The brighter pericardial material is an assumed 0.8 mm shell. No synthetic texture maps were needed for the surrounding anatomy.
Keep the 3D view focused on the heart
Open surfaces are handled as material walls
No re-export is required for this ultrasound prototype. Temporary caps classify interiors; they do not close vessel lumens with solid acoustic plugs. The original FBXs remain unchanged.
| Structure | Objects | Open objects | Treatment |
|---|---|---|---|
| Bones / cartilage | 298 | 0 | Closed interiors; bright interfaces and strong bone attenuation. |
| Liver | 1 | 0 | Closed scattering volume. |
| Blood vessels | 14 | 14 | Temporary lumen closures and approximately 1 mm two-sided walls; lumens stay hollow. |
| Muscles / diaphragm | 26 | 8 | Closed muscle volumes; open sheets use approximately 2.5 mm thickness. |
| Skin | 1 | 1 | Temporary body closure plus an approximately 2 mm skin shell. |
| Lungs / pleura / airway | 5 | 5 | Aerated interiors; approximately 0.7 mm pleural sheets. The airway proxy handles seven nonmanifold edges. |
| Animated pericardium | 1 | 1 | 0.8 mm closed material wall, hollow sac, 28 evaluated poses; source boundary rims are joined. |
Thicknesses are modelling assumptions. The pericardial material wall is watertight and consistently wound; the anatomical openings and sac stay hollow. Thin branches and sheets remain limited by context resolution.
Reference transfer and cycle balance now pass
The earlier fields carried only a small fraction of their prescribed valve flow. A shared face value now gives matching normal velocity from either side. Both circuits use a 54.52 mL stroke target, with balanced atrial storage and explicit zero-storage great-vessel conduits.
| Valve | Upstream mL/s | Downstream mL/s | Target mL/s |
|---|---|---|---|
| Mitral | 0.999974 | 0.999974 | 1.000000 |
| Tricuspid | 0.999979 | 0.999983 | 1.000000 |
| Aortic | 1.000004 | 1.000007 | 1.000000 |
| Pulmonary | 0.999978 | 0.999982 | 1.000000 |
Native WGSL sampling just to either side of every shared reference face. Worst transfer error: 0.0026%. This percentage describes a numerical interface check, not physiological accuracy.
Ejection is gated to the main contraction interval. The spurious outflow pulse between filling peaks is removed. The unchanged healthy-reference ECG remains synchronized to the source cycle.
What remains unresolved
The balanced flow model is not yet conservative on the moving anatomical blood domain. The approximate transported mask obstructs some of the prescribed transfer. The table reports the fraction landing at mask-clear sample points over 112 phases, weighted by prescribed flow.
| Valve | Transfer at mask-clear points |
|---|---|
| Mitral | 56.4% |
| Tricuspid | 67.4% |
| Aortic | 62.6% |
| Pulmonary | 65.3% |
Forcing the same stroke volume through only those inferred openings generated implausibly narrow, fast jets. That experiment was rejected. Credible clinical velocities need verified valve orifices and a moving-domain fluid solve; rendering more particles would not establish that.
The thorax is static, so breathing and lung sliding are absent. Acoustic materials are assigned. Full lung reverberation, rib refraction and coherent Doppler acquisition remain future work. Some apical and oblique presets have poor acoustic windows and are not verified patient probe positions.
Methods and evidence
Native shader checks passed with no WebGPU validation errors. UI control references and module syntax were checked; the live browser interface was not exercised here. The models and recordings have not received clinical review.
Heterogeneous acoustic maps: k-Wave example. Explicit circulation boundary conditions: SimVascular documentation. Clinical colour reference and acquisition discussion: earlier Doppler review. These inform the design; this is an independent simplified implementation.