Hearing the heart: acoustic biomarkers at the point of care

Subtle changes in cardiac sound can flag risk long before symptoms surface. Capturing that signal reliably in a busy clinic — not just a quiet lab — is the real engineering challenge.

Biomarkers hiding in plain sound

Murmurs, timing shifts, and spectral changes carry clinically meaningful information. The difficulty is separating them from ambient noise and from natural variation between patients.

Our models are trained on diverse, real-world recordings so they remain robust across stethoscopes, rooms, and body types rather than overfitting to pristine conditions.

Designed for the point of care

A screening tool only helps if clinicians can trust it in seconds. We surface a calibrated confidence score, not a black-box verdict, so the result fits naturally into an existing workflow.

The stethoscope’s unfinished promise

For two centuries, the stethoscope has let clinicians listen to the heart, and for two centuries the interpretation of what they hear has remained a deeply human, deeply variable skill. A subtle murmur, an extra heart sound, a faint click — these carry real diagnostic information, but extracting it reliably depends on training, attention, and acoustic conditions that vary from clinic to clinic. The signal has always been there. What has been missing is a consistent way to read it.

Acoustic biomarkers aim to finish that promise. The heart is a mechanical pump, and like any pump it broadcasts the state of its valves, chambers, and flow through sound. By capturing that sound digitally and analysing it with models trained on labelled clinical data, we can surface patterns that are difficult for the unaided ear to catch and impossible to document objectively.

What the heart is telling us

A healthy heartbeat has a clean, repeating structure: the closing of valves produces the familiar pair of sounds, separated by intervals that reflect the timing of the cardiac cycle. Disease distorts this structure in characteristic ways. Turbulent flow across a narrowed or leaking valve generates murmurs whose timing, shape, and frequency content point toward specific conditions. Stiffened chambers add extra sounds. Irregular rhythm scrambles the timing itself.

Each of these distortions is an acoustic biomarker — a measurable feature of the sound that correlates with an underlying physiological state. The diagnostic art lies in distinguishing a benign flow murmur from a pathological one, or an incidental extra sound from a sign of failure. These distinctions hinge on fine details of timing and spectral shape that are exactly the kind of pattern a well-trained model excels at.

Capturing a clean signal at the bedside

Point-of-care recording is harder than it sounds. The chest is a noisy place: breathing, muscle movement, ambient room sound, and the rustle of the sensor against skin all compete with the heart. Placement matters too, since different listening positions emphasise different valves. Our capture protocol guides the operator to consistent positions and uses signal-quality checks to reject recordings too contaminated by noise to analyse.

Once a clean recording is in hand, the model segments it into individual cardiac cycles, aligning the heart sounds so that murmurs and extra sounds can be located precisely within the cycle. This segmentation is the foundation for everything that follows: a murmur in early systole means something different from one in late diastole, and only by anchoring events to the cycle can the model reason about them correctly.

From sound to actionable flag

The output is designed for the realities of frontline care. Rather than an opaque label, the system reports a calibrated likelihood of clinically significant findings, along with the segment of audio that drove the conclusion. A clinician can listen to exactly the heartbeat the model flagged, building trust rather than demanding blind faith. In a screening context, this turns a brief bedside recording into a triage decision: reassure and move on, or escalate to imaging and specialist review.

Why point of care changes the math

The power of acoustic biomarkers is not that they replace echocardiography or specialist assessment — they do not. It is that they bring a first, objective screen to places and moments where those resources are absent. A primary care visit, a rural clinic, a community health screening: in all of these, a low-cost recording analysed on a phone or tablet can identify the patients who genuinely need to travel for advanced imaging, and spare the many who do not.

This redistribution of attention is where the real impact lies. Specialist cardiology capacity is scarce and unevenly distributed. By pushing a reliable first screen out to the point of care, acoustic biomarkers concentrate that scarce capacity on the patients most likely to benefit, and catch conditions earlier in people who might otherwise never have been flagged at all.

The heart has always been speaking. With consistent capture and calibrated interpretation, we can finally listen at scale — turning a centuries-old clinical ritual into a repeatable, objective measurement available far beyond the specialist’s office.