Classification of Atrial Fibrillation ECG Signals Using 2D CNN

The paper evaluates the significance of the extracted ECG signal features to classify and diagnose Atrial Fibrillation (AF) using a 2D Convolutional Neural Network (CNN). The input to the network is a 2D scalogram calculated by applying the Continuous Wavelet Transform (CWT) on five heartbeats segments of an ECG signal. The motivation behind converting the ECG signal to a 2D image is to use deep features, including time and frequency domain features, contained in the image, rather than the raw ECG data for training purposes. The dataset used for this research was obtained from the MIT-BIH Atrial Fibrillation (MIT-BIH-AF) database. In this paper, we have identified the best hyperparameters for the 2D CNN model that could successfully predict AF with an accuracy of 99.12%. The paper showed that it is possible to observe all significant changes in the signal using a scalogram in the AF classification using 2D CNN.