A New Atrial Fibrillation Detection System with Noise Cancellation and Signal Annotation

This paper describes a new heart arrhythmia detection system, with an emphasis on the Atrial Fibrillation (AF) detection algorithms. This system is designed to record lead-II ECG signal for several days, which makes it suitable for digital health monitoring and AF detection. The algorithm consists of three stages. In the first stage, data is preprocessed to reduce the high-frequency noise as well as Baseline Wandering (BW) of the signal using a novel method. In the second stage R-peaks of the ECG signal are detected. At last, Heart Rate Variability (HRV) features are extracted and AF detection is carried out using a low-complexity Decision Tree (DT) method. In addition, three Feature Selection (FS) methods are performed on HRV features, where the optimal combination of them is selected for the classification task. The algorithm is implemented and verified using three publicly available databases as well as the practical data taken from the designed ECG device. The results are comparable to the state-of-the-art algorithms by using only HRV features and DT. This makes the algorithm suitable for real time and low power applications, which are the requirements of the wearable ECG patch for the digital health.