FDA Clears Belfast AI That Reads the Smartwatch ECGs Apple Watch Can't

A Belfast startup says it has solved the problem that kept smartwatch heart monitors out of the clinic. PulseAI announced Wednesday that the FDA cleared its AI-powered arrhythmia model under the 510(k) pathway — to TechTimes' knowledge, the first cloud-based deep neural network specifically cleared to interpret wearable single-lead ECGs. In its pivotal study of 296 participants, published in Heart Rhythm, the algorithm cut inconclusive readings by 92 percent, dropping the inconclusive rate from roughly 20 percent to under one percent.

That 20 percent figure is the heart of the problem. The Apple Watch records 30 seconds of electrical signal and returns one of three answers: atrial fibrillation, normal sinus rhythm, or nothing useful. About one in five times, it lands on inconclusive. Single-lead ECGs see one electrical view of the heart instead of the standard twelve, and noisy or motion-contaminated waveforms leave on-device algorithms unable to commit. An inconclusive result isn't free — it triggers repeat attempts, physician calls, Holter monitors and office visits, with documented cases running up thousands of dollars in testing and no change in care.

PulseAI's fix is to move the thinking off the wrist. The watch still records; the waveform gets transmitted to the company's servers, where its proprietary PDNN — trained on more than one million patient ECG recordings — interprets it and sends back a structured rhythm classification. No new hardware, no different behavior from the wearer. The cloud just has more room to work than a battery-constrained watch processor does.

The clinical data behind the clearance came in stages. A prospective study of 91 patients at Beacon Hospital in Dublin, presented at the Heart Rhythm Conference in 2023, logged 93 percent sensitivity and 96 percent specificity for AFib detection with an inconclusive rate below one percent — while the Apple Watch's native algorithm, run on the same recordings, managed 76 percent sensitivity, 85 percent specificity and a 12 percent inconclusive rate. A second study at University Hospital Basel, led by Dr. Patrick Badertscher and published in Cardiovascular Digital Health in 2024, added European validation.

Worth keeping the clearance in perspective: the FDA doesn't independently test devices under 510(k). It reviews the manufacturer's data and decides the device is substantially equivalent to a predicate — not that it's better. But for Software as a Medical Device, clearance is the line between a wellness app and a product hospitals can actually buy, with a path toward reimbursement and the right to market as an FDA-cleared medical device.

The timing lands in a hot stretch for cardiac AI. The FDA cleared 295 AI-based medical devices in 2025 from 221 manufacturers, 26 of them cardiovascular — the second-biggest category after radiology — with a median clearance time of 142 days. PulseAI joins a cohort that includes AccurKardia, whose AccurECG 2.0 platform won 510(k) clearance in January 2026 for near-real-time ECG interpretation in clinical settings. The two are chasing different rooms: AccurKardia goes after hospital and enterprise ECG workflows, while PulseAI aims at the wrist. With AFib affecting an estimated 33 million people worldwide and carrying a fivefold higher stroke risk, the stakes are population-sized.