Automated detection of atrial fibrillation episode using novel heart rate variability features.

Atrial fibrillation (AF) is one of the most common life-threatening arrhythmia affecting around six million adults in the US. Typical detection of AF requires tedious and manual analysis of ECG which can often delay medical intervention. With the advent of wearable devices that can accurately record the time interval between two heartbeats (RR interval), automated and timely detection of AF is now possible. In this paper, we engineer novel heart rate variability features based on linear and non-linear dynamics of RR intervals. Unlike complex features extracted from ECG signals, these features can be easily obtained using wearable sensors. We propose automated classifiers to detect AF episodes and also compare the performance of different classifiers. Our proposed classifier has a very high sensitivity (98%) and specificity (95%) and outperforms prior published works.