An AI Stethoscope Missed Most Cat Heart Murmurs

An AI-powered stethoscope, trained mostly on human heartbeats, missed most cat heart murmurs and wrongly told owners of healthy dogs their pets had an irregular heartbeat. That's what a new study from North Carolina State University found. Veterinary students matched the device. Experienced vets beat it outright.

KEY NUMBER
The AI stethoscope correctly caught only 2 of 22 real heart murmurs in cats, while telling 22 healthy dogs they had a heart rhythm problem they didn't have.

What Happened

Cardiologists at NC State's veterinary teaching hospital examined 54 dogs and 51 cats. They used an AI-enabled digital stethoscope, one that listens to the heart and makes its own call on murmurs and rhythm problems. A board-certified cardiologist, a cardiology resident, and a fourth-year veterinary student each examined every animal too, giving the team a human baseline to compare against.

In dogs, the AI caught 33 of 38 real heart murmurs, matching the vet students exactly. But it flagged every single dog as having some kind of irregular heartbeat, even ones with a perfectly normal rhythm. Of 28 dogs it called atrial fibrillation, 22 of those calls were wrong. In cats, the AI caught only 2 of 22 real murmurs. Veterinary students, without any AI help, correctly spotted nearly two-thirds of them on their own.

Why It Matters

The AI behind these stethoscopes is trained mostly on human heart sounds, not cat or dog hearts. A cat's heart beats faster and sounds different from a human's, or even a dog's, and this study suggests that gap matters more than the marketing for these devices lets on. Some veterinarians told the researchers they'd started second-guessing their own exams because the device disagreed with them, exactly the kind of overreliance a mislabeled "universal" tool can cause. The same false-alarm problem shows up in human hospitals, where machine learning has been tested specifically to cut false alerts without missing real emergencies.

What Happens Next

The researchers want makers of these devices to test and label them separately for each species, rather than selling one "universal" AI model trained mainly on humans. Until then, the study's authors say the tool should support a vet's own exam, not replace it, especially for cats.

Takeaway

The stethoscope excelled at exactly the population it was trained on: human hearts, and dog hearts that sound closer to them. It struggled the moment the anatomy changed. That's the same test any AI diagnostic tool eventually faces. Performing well on familiar data says nothing about performing well once the population does.