Archived version 6 audit. Reference transfer, valve windows and cycle balance have since been revised. Read the current correction and thorax review →
FLOW & DOPPLER AUDIT · 8 SEPTEMBER 2026
Recognizable Doppler. Significant flow errors.
The direction convention works, and the view shares broad visual cues with normal echo. Deeper measurements found failed valve transfer, unequal left/right output and an extra outflow pulse. The audited flow could not be described as a validated healthy circulation.
Audited: version 6, at 64.3 bpm. This update adds validation findings; it does not retune the recorded flow. Measurements cover 8,192 cycle samples, four valve interfaces, 28 phases in two scan planes and eight new native WebGPU renders.
What passes, and what fails
| Check | Result | What it means |
|---|---|---|
| Beam direction and aliasing equation | Software checks pass | Red toward / blue away, beam-perpendicular zero and Nyquist wrapping behave consistently. This does not validate the input velocities. |
| Predominant flow direction | Plausible at selected phases | Before aliasing, 96.9% of visible sampled A4C flow points toward the probe during early filling; 87.0% of sampled A5C flow points away during main ejection. |
| Transfer across valve boundaries | Fails | A prescribed 1 mL/s transfer produces only 0.014–0.057 mL/s across the reference-grid interfaces, with different values on their two sides. |
| Whole-cycle circulation balance | Fails | Prescribed LV output is 72.88 mL per cycle versus RV 55.43 mL: 31.5% higher. Great-vessel volume changes are also unbalanced. |
| Valve timing | Fails | A secondary LV volume decrease creates 13.64 mL of extra outflow between filling peaks, outside the main contraction interval. |
| Jet texture and vortex evolution | Partial resemblance | Colour is recognizably Doppler-like, but conspicuous bands, simple grain and a fixed recirculation shape remain. Moving particles do not prove realistic blood transport. |
| Absolute velocity / clinical E/A | Unvalidated | There is no patient-matched flow reference or spectral Doppler estimator. Volume-flow peak ratios are not E/A velocity measurements. |
Direction percentages use coarse central-plane samples and a moving-tissue visibility mask, not final image pixel counts or clinical accuracy scores. They are measured at phases 57.1% and 17.9%, respectively.
The circulation needs correction before visual tuning
The code routes every decrease in ventricular volume to outflow. The inferred LV volume falls a second time between early and late filling, producing another aortic pulse. This is visible in both the source-frame volume estimates and the fitted curve; smoothing alone is not a demonstrated fix.
The stored velocity fields are solved inside separate closed chamber labels. They use internal sources and sinks, without enforcing the required transfer across each shared open valve. The independent face integration below was confirmed using the actual WGSL sampler.
This is a reference-grid continuity test with unit flux, zero recirculation and no wall transport. It is not a measurement of anatomical valve area or a clinical accuracy percentage. A small residual from the original potential solve does not establish continuous valve flow.
Inspect the Doppler appearance
The normal reference clips show focused colour regions and changing inflow/outflow patterns. Our renders reproduce broad direction cues, but the valve region has strong contour-like alias bands, and the tissue and colour boundaries are unusually clean. This is a qualitative comparison with different anatomy and acquisition settings.
Images use the actual anatomy and acoustic shaders, rendered offline with Dawn / SwiftShader. Phase text, ROI outline and colour bar were added afterward. They are not browser screenshots. All capture settings are listed in the downloadable GPU measurements.
Other A4C phases: main ejection and the interval between filling peaks
Colour scale comparison: ±34.2 versus ±51.3 cm/s
The default 4.5 MHz / 4 kHz setting gives ±34.2 cm/s. BSE’s initial setting for routine regurgitation colour assessment is 50–60 cm/s; a lower scale can be appropriate for other tasks. At the selected early-filling and ejection samples, 31.2% and 45.0% exceed our default limit before wrapping. BSE minimum dataset ↗
Frequency currently affects both B-mode and colour. At 4.5 MHz and 150 mm depth, the implemented round-trip PRF limit permits only about ±43.9 cm/s. A separate Doppler frequency would allow more useful acquisition settings without reducing B-mode frequency.
Apical motion still needs a physiological reference
The particles circulate more visibly, but the added ventricular roll has a fixed shape and a changing amplitude. Healthy 4D-flow observations show a developing early-filling vortex that deforms and moves toward the apex, often followed by a new late-filling ring nearer the mitral valve. Our current field does not reproduce that evolution. Elbaz et al., healthy 4D-flow observations ↗
Particles also slide, reject steps and reseed near boundaries. Their accepted displacement can differ from the velocity shown by Doppler. They cannot currently measure true blood residence time, washout or stagnation.
What to correct next
Verify cavities and valves through the animation. Reconcile the volume estimates, left/right output and actual valve opening phases. The secondary volume decrease needs a cause, not a cosmetic speed adjustment.
Enforce connected flow and valve timing. Use a conservative moving blood-domain model with explicit valve transfer and wall motion; check both sides of each valve and whole-cycle balance.
Model evolving jets and vortices. Fit a documented healthy case, then compare another measured case independently.
Improve Doppler acquisition. Separate colour frequency from B-mode and add coherent ensemble timing, beam smoothing and clutter behaviour. Retain adjustable Nyquist and wall filtering.
Scope and references
The real Doppler clips, MRI observations and ECG recordings are independent examples. There is no synchronized patient-matched dataset for this atlas, and no sonographer or clinical review has been performed. Native checks completed without GPU validation errors; that establishes execution, not healthy haemodynamics.
Aquilina et al. (2007) — normal apical echo clips, Figures 11 and 14; linked at source, not redistributed (CC BY-NC-SA 3.0). Robinson et al. / BSE (2020) — routine colour acquisition guidance. Elbaz et al. (2014) — measured healthy vortex evolution (CC BY 4.0).