1.73 → 0.25 dB
Same murmur, two chest walls.
Synthetic grade 2 murmur through modelled 60 mm and 110 mm chest walls, before and after correction. n = 600. Bench.
3.7 mm
RMS error, wall thickness.
Estimated from the take against a measured wall, 111 recordings. Bench.
6.0 dB
Noise rejected.
Reference capsule against a clinic room recording. Bench, one room, three assemblies.
45%
Contact threshold.
Below it, no result. The gate was set on bench takes with deliberate poor contact.
0
Nothing fitted to a population.
The correction is a division. No training set, so no inherited bias and no weights to audit.
It does not find disease, grade a murmur, or replace an echo.
It corrects a recording for the body it was taken through. A corrected recording is still a recording. The next step is a pilot with echo-booked patients.
What the pilot will measure.
Twenty to forty patients with an echo already booked, recorded before the echo is read. The pilot produces the number the bench cannot: agreement with the reference the clinic already uses.
Agreement
Whether the corrected level tracks the echo grade across body types.
Repeatability
Two operators, same patient, same room. The variance that matters day to day.
Refusals
How often the gate says no, and whether those refusals were right.
Time
Added minutes per patient, measured rather than estimated.