Pattern recognition and automatic identification of early-stage atrial fibrillation

摘要

Atrial fibrillation (AF) is a common cardiac arrhythmia and is responsible for a number of complications. While early-stage AF typically lasts only a few episodes and may not be immediately life-threatening, the cardiac arrhythmia favors electrical and structural alteration of the atria that tends to intensify and perpetuate AF even at incipient stage. Therefore, recognition and identification of patterns associated with early-stage AF episodes is demanded for effective treatment and disease management. Nonetheless, the brevity of early-stage AF negates a myriad of conventional models for effective detection, particularly when only a few seconds of recording is available. In this paper, we investigate constructive patterns based upon intrinsic time-scale decomposition (ITD) to parse single-lead ECG signals with short duration collected from wearable devices. ITD provides accurate instantaneous time-frequency-energy characteristics for the nonlinear and nonstationary data, particularly the short-term time series signals. Our model registers average accuracy of 95%, specificity of 96% and sensitivity of 93% for the diagnosis of AF events, handily outperforming wavelet-based algorithm. (C) 2020 Elsevier Ltd. All rights reserved.