Atrial Fibrillation Detection from PPG Interbeat Intervals via a Recurrent Neural Network

Abstract

Atrial fibrillation (AF) affects millions of individuals worldwide and can lead to serious complications such as stroke or heart failure. This arrhythmia is difficult to diagnose with ambulatory electrocardiogram monitors in the early stages due to its transient nature. Recent advances in wearable photoplethysmographic (PPG) devices are promising for screening AF in large populations as they are relatively comfortable and can be worn over long periods of time. Herein, we propose a system to detect AF from PPG recordings. This system is composed of a beat detector to extract interbeat intervals and a classifier for detection. We trained the classifier on a large public database of interbeat intervals and then evaluated the whole system on PPG recordings collected during catheter ablation procedures. We achieve an accuracy of 0.986 for the detection of AF with a sensitivity and specificity of 1.0 and 0.978 respectively. These metrics compare favorably with existing systems.