Medical Technology · us
Fewer “Unclassifiable” Smartwatch ECGs: PulseAI Arrhythmia Analysis Software Receives FDA Clearance
Wearable devices can record ECGs, but they cannot always provide an answer. PulseAI has received U.S. marketing clearance, and a study of 296 participants also showed that its algorithm could reduce inconclusive results. Whether the study’s performance extends to different devices and routine care, however, remains to be validated.
When an ECG recorded on the wrist returns only an “unclassifiable” result, patients and care teams still face repeat measurements, manual review, and follow-up. On November 5, PulseAI announced that its wearable arrhythmia analysis software had received 510(k) clearance from the U.S. Food and Drug Administration (FDA), aiming to make these everyday measurements easier to translate into information that can inform clinical review.
According to FDA clearance documents, the product, named PulseAPI SmartWatch, was cleared on October 27, 2026, as a Class II medical device, with AliveCor’s KardiaAI V2 serving as the predicate device for substantial equivalence. This regulatory determination confirms that the product meets the marketing requirements of that pathway; it does not mean the FDA has determined that it is superior to all existing ECG analysis tools.
The software analyzes portable ECG recordings from adults under a physician’s care. The company says recordings are sent to the cloud for analysis, with structured results returned to an integrated application for clinical review. Compatible hardware listed in FDA documents includes Apple Watch Series 6, Samsung Galaxy Watch, Fitbit Sense, Withings ScanWatch, and specified AliveCor devices, but the supported rhythm classifications vary by hardware. The product is also not intended for patients with implanted electronic devices or for use as an alarm system.
The evidence cited by the company is a prospective observational study published online in Heart Rhythm in November 2025. According to Ulster University’s research record, the study was conducted at a teaching hospital in Ireland and included 296 participants after excluding those who did not meet the eligibility criteria. The same Apple Watch ECG recordings were interpreted separately by the native algorithm and an AI model to compare their performance in identifying atrial fibrillation. Using the same recordings helps reduce the influence of measurement differences on the comparison.
The study also addressed a question that accuracy figures can easily obscure: how should unclassifiable results be counted? The team evaluated performance by excluding inconclusive results, treating them as interpretation errors, and allowing up to three measurements. These choices change the results and also show why looking only at recordings that receive a classification may overestimate a tool’s performance in actual use.
When up to three measurements were allowed, the AI’s sensitivity for identifying atrial fibrillation was 98.4%, and its specificity was 96.6%; the corresponding figures for the native Apple Watch algorithm were 92.2% and 91.0%. Sensitivity reflects the ability to detect atrial fibrillation, while specificity reflects the ability to correctly identify its absence. The AI produced a relative reduction of 92% in unclassifiable results, but this figure does not mean that overall accuracy increased by 92 percentage points.
Reducing inconclusive results may allow care teams to spend less time handling unclassified recordings. This study cannot yet answer whether that reduces strokes, improves long-term health outcomes, or actually lowers the care workload. PulseAI was among the study’s funders, and the data came from a single hospital using Apple Watch recordings, so the same performance cannot be directly assumed for all compatible hardware. Following clearance, validation across devices and routine care settings will determine how much this tool can help.