Ensemble Learning for Atrial Fibrillation Screening from a Single Lead ECG Wave of Wearable Devices

The advancements of wearable devices allow for collecting physiological signals (such as single-lead ECG) continuously, and mining data from these physiological signals has become a research hot. Atrial Fibrillation (AF) is one of the most common cardiac arrhythmias, and effective AF detection would avoid serious complications, like stroke and coronary artery disease. This paper presents an ensemble learning-based algorithm for automatic AF detection from a single-lead ECG recording, which consists of data preprocessing, image transformation and ensemble learning. The data preprocessing stage is used to tackle the noise and disturbance problems. Afterward, the ECG recordings are converted into time-frequency spectrums and Poincare plots. Then, the ensemble learning architecture can obtain the final classification result. Experiments results demonstrate the effectiveness of the proposed algorithm.