Atrial Fibrillation Dynamic Features Recognition using 1-D Convolutional Neural Networks

Atrial fibrillation (AF) is a prevalent cardiac arrhythmia that requires accurate detection for effective management and treatment. This study aims to propose a simple yet effective one-dimensional convolutional neural network (1D-CNN) for the AF dynamic features recognition which are extracted using a second-order dynamic system (SODS) concept. The SODS concept utilizes a dynamic system approach to capture the short-term behavior of ECG signals. It employs a second-order differential equation to model the ECG signals based on parameters such as the natural frequency (ω), damping coefficient (ξ), and forcing input (u) to enable the characterization and differentiation of ECG signal. The 1D-CNN model was trained and validated with preprocessed ECG signals from the MIT-BIH Atrial Fibrillation and MIT-BIH Normal Sinus Rhythm database. The results demonstrate the efficacy of the SODS and 1D CNN method for AF recognition. The proposed approach achieves high accuracy, with an accuracy of 98.35% using natural frequency, ω feature.