A dog's heart beats roughly 60 to 140 times a minute. A specialist cardiac workup at a veterinary referral hospital can cost $800 to $1,600. For many owners those two facts collide at exactly the wrong moment, in a waiting room with a pet who has started coughing at night and a vet who has heard something faint and wants to send you to a cardiologist you cannot easily afford or schedule.
That gap is the market opportunity a Cambridge startup called Sonus Health is trying to close. Their app turns any smartphone into a preliminary cardiac screening tool: hold the phone microphone against the pet's chest for about thirty seconds, get an instant reading of heart rate, heart rate variability, and indications of possible murmurs, then a full report reviewed by a board-certified veterinary cardiologist returns within 48 hours.
It is a real product with real data. It also is not a replacement for a vet. Both of those statements matter, and most marketing on this stuff blurs the line between them.
What the research actually shows
The underlying machine learning was published in January 2026 as Jose et al. on arXiv (paper 2601.13593), "Instant Preliminary Cardiac Analysis from Smartphone Auscultation." The dataset: more than one hundred recordings from dogs across four continents, with 38 of those recordings annotated by board-certified veterinary cardiologists for evaluation.
The numbers, in order, from the abstract. The primary sixty-second model achieved 91.63 percent mean heart rate accuracy, with 94.95 percent at the median. A faster model optimized for 30 to 40 second recordings achieved 88.86 percent mean, 92.98 percent median. The system uses a multi-stage fallback architecture with quality-aware filtering, which means it can flag recordings it cannot interpret reliably rather than forcing a guess.
The product is built by Decorte Future Industries, a UK startup founded by a University of Cambridge PhD. The company has a 150-plus clinic customer base and ships in more than six continents. Pricing starts at $15.99 per month for owners, with the equivalent specialist assessment valued externally at $400 to $1,100.
Why screening-grade is not diagnostic-grade
This is where I want to be direct with pet parents, because the distinction gets flattened in advertising.
A 91.63 percent mean heart rate accuracy is a specific claim about a specific signal. It means the model is reasonably good at counting beats per minute on a cooperative dog in a reasonably quiet room. It is not the same as a 91.63 percent diagnostic sensitivity for heart disease. It is not a substitute for echocardiography, which is the actual tool your cardiologist uses to evaluate structural cardiac problems. Sonus makes this distinction explicitly. Every full report loops in a board-certified veterinary cardiologist for final review, and the company's own messaging places the product as a filter that decides when a specialist referral is worth the cost and not as a replacement for that specialist.
That is a defensible design. A recent peer-reviewed audit of commercial veterinary AI (Brundage, 2026, DOI: 10.3389/fvets.2026.1761038) found that only 15.8 percent of imaging AI vendors and zero percent of generative AI vendors disclose confidence intervals for their performance claims. The same audit found that only one vendor in 71 disclosed the signalment distribution of their training data. In that context, "we publish accuracy on an arXiv paper with cardiologist-annotated data" is an unusually high-transparency baseline for this market.
What is and is not actually new here
Smartphone cardiac monitoring for dogs is not a 2026 invention. A 2019 PMC feasibility study from Sony's Xperia research team demonstrated that a smartphone-only mechanocardiography pipeline could estimate heart rate for resting dogs in a home setting, with the authors noting that motion artifacts and panting are the primary signal quality limitations. That limitation has not disappeared in 2026. It has been mitigated through better signal processing and a human-in-the-loop reviewer, not eliminated.
Independently, a recent October 2025 study in Animals (DOI: 10.3390/ani15213081) by Brady and colleagues at Dublin City University showed that a collar-mounted inertial sensor and on-device machine learning could automatically classify a specific trained behavior with 92 percent accuracy under a leave-one-dog-out protocol. Different problem, different hardware, but same underlying pattern: small sensors plus lightweight ML plus a clear clinical workflow are producing genuinely useful screening tools for pet health. The category is maturing.
What I would actually recommend to a pet parent weighing this
Three questions before paying for a subscription.
Does your dog tolerate the recording environment? Accuracy drops sharply with motion artifacts, panting, and ambient noise. If your dog will not hold still for thirty seconds indoors, this is a cat-toleration problem and not a technology problem. Pancake will tolerate almost anything. Gigi will tolerate it for about twenty minutes before looking deeply betrayed. Roger will simply refuse. Every household has a Roger.
Do you have a vet you trust who can act on the output? A preliminary murmur flag without a follow-through relationship with a primary care veterinarian is a source of anxiety rather than a source of care. This tool works best as an input into an existing relationship, not a substitute for one.
Are you using this to replace a specialist visit or to decide whether you need one? The honest answer, for a healthy adult pet with no symptoms, is usually "to decide whether you need one." That is the use case the validation data actually supports. If your vet has already heard a murmur and wants an echocardiogram, a smartphone screening tool is not the right next step.
Screening-grade health tech has a real and growing place in responsible pet care. It does not replace the clinic. It does change the economics of when you go and why.
