Diagnostic accuracy of an artificial intelligence-enhanced pulse oximeter for atrial fibrillation detection: a real-world population study

PurposeFingertip-based photoplethysmographic (PPG) devices offer a noninvasive method for at-home atrial fibrillation (AF) screening. In this study, we evaluated the accuracy of a fingertip pulse oximeter enhanced with a supervised artificial intelligence (AI) algorithm in diagnosing AF in a real-world setting.Materials and methodsThis prospective, multicenter study included 401 individuals who underwent concurrent assessment using an AI-enhanced fingertip pulse oximeter and standard 12-lead electrocardiography (ECG), with expert-interpreted ECG serving as the diagnostic reference standard. The oximeter measured pulse rate, oxygen saturation, perfusion index (PI; a PPG-derived indicator of peripheral perfusion), and PPG waveforms, which were analyzed with an embedded AI module to classify heart rhythm as “normal,” “irregular,” or “possible AF.”ResultsWhen benchmarked against expert-interpreted ECG, the AI-enhanced pulse oximeter demonstrated high diagnostic performance for AF detection, with a sensitivity of 96.7%, specificity of 90.5%, positive predictive value of 91.9%, and negative predictive value of 96.1%. AF detection rates derived from the AI-enhanced pulse oximeter were consistent with those identified by computer-read 12-lead ECG (55.4% vs. 52.6%, p = 0.436). Diagnostic performance remained high across all clinical subgroups, with a minor decrease observed among individuals with a perfusion index value of ≤1%. No severe device-related adverse events were reported.ConclusionsThe portable, site-less design of this fingertip-based PPG device incorporating an AI algorithm supports its scalability for remote, real-time rhythm monitoring and population-level AF screening.