Towards Early Detection and Burden Estimation of Atrial Fibrillation in an Ambulatory Free-living Environment

Abstract

Early detection and accurate burden estimation of atrial fibrillation (AFib) can provide the foundation for effective physician treatment. New approaches to accomplish this have attracted tremendous attention in recent years. In this paper, we develop a novel passive smartwatch-based system to detect AFib episodes and estimate the AFib burden in an ambulatory free-living environment without user engagement. Our system leverages a built-in PPG sensor to collect heart rhythm without user engagement. Then, a data preprocessor module includes time-frequency (TF) analysis to augment features in both the time and frequency domain. Finally, a lightweight multi-view convolutional neural network consisting of 19 layers achieves the AFib detection. To validate our system, we carry out a research study that enrolls 53 participants across three months, where we collect and annotate more than 27,622 hours of data. Our system achieves an average of 91.6% accuracy, 93.0% specificity, and 90.8% sensitivity without dropping any data. Moreover, our system takes 0.51 million parameters and costs 5.18 ms per inference. These results reveal that our proposed system can provide a clinical assessment of AFib in daily living.