Sonorus uses AI to 'listen' for disease

Rheumatic heart disease often starts with an ordinary sore throat. A group A streptococcal infection can trigger acute rheumatic fever, and repeated attacks can scar the heart valves years before a child looks obviously sick. (who.int) Heart valves are the flaps that keep blood moving one way, like doors in a subway station. When those doors get stiff or leaky, blood flow turns turbulent, and that turbulence can create a murmur a clinician can hear through a stethoscope. (nhlbi.nih.gov) The standard way to catch early rheumatic heart disease is echocardiography, which is ultrasound for the heart. It works well, but it needs equipment, trained operators, and follow-up systems that are often missing in the places where rheumatic heart disease is most common. (thelancet.com00127-X/fulltext)) That is the gap Sonorus is trying to squeeze into. The Australian startup says it is building a tool that records heart sounds and uses artificial intelligence to flag signs of rheumatic heart disease within minutes instead of sending every child straight to imaging. (mashable.com) The basic bet is that a cheap sound signal can be used as a front-door screen. Sonorus describes its product as a low-cost, portable mass-screening and triage tool for communities at risk, not as a replacement for a full heart scan. (mashable.com, sonorus.tech) That distinction matters because rheumatic heart disease is concentrated in poorer and more remote settings. Sonorus says its team formed around health inequity in Australia, while Monash University’s profile of the company says the disease causes about 200,000 to 250,000 premature deaths a year and hits children and young people especially hard in low- and middle-income countries. (sonorus.tech, monash.edu) The company is still early. Sonorus says it is in the data-collection stage of a plan to screen 1,000,000 people, which means the hard part now is building a training set large enough for the model to learn what normal and abnormal heart sounds really look like across different ages and settings. (sonorus.tech) Outside descriptions of the prototype show why people are paying attention but also why caution is needed. Prototypes for Humanity says Sonorus analyzes heart sounds to detect murmurs with 80 percent specificity, which is promising for an early screen but not the same thing as a definitive diagnosis. (prototypesforhumanity.com) This is not a one-company idea. Research groups and medical societies are already exploring two related shortcuts: using artificial intelligence to help nonexperts capture echocardiography images, and using machine learning to classify abnormal heart sounds from digital stethoscopes. (onlinejase.com00133-5/fulltext), ieeexplore.ieee.org) If this works, the workflow changes from “find an ultrasound machine first” to “listen first, scan second.” In places with too few cardiologists and too many children to screen, that could mean a nurse with a digital stethoscope decides who most urgently needs the scarce imaging slot. (mashable.com, thelancet.com00127-X/fulltext)) The catch is that rheumatic heart disease is a valve disease with formal imaging criteria, so sound alone has to prove it can generalize beyond a demo and hold up in real clinics. Until that happens, Sonorus is best understood as part of a broader push toward cheaper diagnostic signals, where a few seconds of audio might decide who gets the expensive test next. (world-heart-federation.org, mashable.com)