A new study reveals that consumer-grade AI stethoscopes cannot reliably replace human ears in veterinary clinics.
Can an algorithm listen to a beating heart better than a veterinary student? Tech companies want us to believe AI-enabled stethoscopes can democratize diagnostics. But a new trial reveals a stark reality: the technology is barely keeping pace with novices and completely misses the mark on feline patients.
This challenges the narrative of plug-and-play clinical AI. If an AI tool cannot outperform a fourth-year student and completely fails on cats, it is not ready for solo clinical duty. It risks creating a false sense of security for pet owners and over-burdening clinics with false positives.
The canine diagnostic gap
In dogs, the Core 500 stethoscope achieved a sensitivity of 86.8% and a positive predictive value of 82.5% for detecting murmurs. However, its specificity crawled at just 56.3%. Its agreement with experienced clinicians yielded a kappa of 0.447, matching fourth-year veterinary students exactly. The AI is not a super-expert. It is a digital student that requires constant supervision to avoid misdiagnoses.
The algorithm’s success also heavily relies on the severity of the condition. High-grade murmurs of grade 3 or higher had an odds ratio of 15.11 for detection. This means the algorithm mostly excels at spotting what humans can already easily hear.
A total feline failure
The feline data is far worse. The AI detected only 2 of 22 murmurs in cats, resulting in a dismal 9.1% sensitivity and a kappa of 0.081. This is a dangerous blind spot for veterinary practices dealing with mixed-species caseloads.
Arrhythmia detection was equally chaotic. While the device caught atrial fibrillation with 100% sensitivity, it also flagged every single dog as having an arrhythmia. It classified zero dogs as arrhythmia-free. A test that calls everyone sick is clinically useless.
Why this matters
This performance gap highlights a persistent issue in veterinary AI. While earlier research on canine cardiac murmurs via digital wireless stethoscope showed promise in controlled settings, real-world clinical environments present a much steeper challenge. If clinicians rely on these tools to screen patients, they face a double threat: missing life-threatening heart disease in cats, while drowning in false-alarm arrhythmia referrals for dogs.
We must stop treating clinical AI as a finished product. It is an assistant that still needs supervision.
- Dog murmur sensitivity: 86.8%
- Cat murmur sensitivity: 9.1%
- Odds ratio for detecting high-grade murmurs: 15.11
- Arrhythmia-free dogs detected: 0%
This prospective trial was conducted at a single university teaching hospital over a five-month period, meaning results may vary in quieter private practices. Read the full findings in the Journal of the American Veterinary Medical Association.