Atrial Fibrillation Detection Using Photoplethysmographic Signals From a Smartwatch

Abstract

Atrial Fibrillation (AF) with high mortality rate needs to be monitored and detected accurately. To record Photoplethysmography (PPG) signal, wrist-watches are used. The PPG signal gets corrupted by motion and noise artifacts during monitoring, and is detected using proposed Motion and Noise artifact (MNA) algorithm based on accelerometer reading and Time-Frequency spectra (TFS) of PPG signal. Since AF is indicated as varying pulse-to-pulse intervals in a PPG signal, AF detection is made using root mean square of successive differences and sample entropy features, discriminating AF from Normal Sinus Rhythm (NSR). The presence of Premature Atrial and Ventricular Contraction (PAC/PVC) in subjects due to its randomness may also lead to false AF detection. The poincaré plot based PAC/PVC detection implemented in this research not only separates PAC/PVC from NSR and AF but also improves the accuracy of AF and NSR detection. The proposed Convolutional Neural Network (CNN) on Time-Frequency spectra (TFS) ultimately proves to give better results for classification of PPG signal into AF, NSR and PAC/PVC. The results for training and testing are validated on University of Massachusetts Medical Center (UMMC) Simband and Medical Information Mart for Intensive Care III databases with higher accuracy, sensitivity and specificity values.