Automatic Identification and Location of Paroxysmal Atrial Fibrillation Based on Single Heartbeat from Dynamic Electrocardiogram

Bailing Zhang, Shaochang Wang, Yi Xin, Ying Zhao

Abstract: Atrial fibrillation (AF) is a prevalent arrhythmia in clinical practice, with potentially serious consequences. Due to the difficulty in capturing most episodes of paroxysmal atrial fibrillation (PAF), there is an urgent need for real-time monitoring of patients wearing long-term electrocardiogram recording devices. This study proposes a method for identifying and locating PAF in long-term ECG recordings, using the CPSC2021 dataset. The first part of the method consists of an identification algorithm based on a 5-minute single-lead ECG segment. Thirty-two features were extracted from the RR interval sequence, and three machine learning models were trained, with support vector machines (SVM) demonstrating the best performance. The second part of the method involves a PAF location algorithm based on single heartbeats. A convolutional neural network (CNN) model was trained to identify whether AF had occurred, and the location score was found to be superior to that of the baseline method given by CPSC2021. The proposed method has potential applications in portable dynamic ECG monitors.