Deep learning is one of the fastest growing areas of machine learning. With improvements in hardware technology, the use of deep learning algorithms is increasing daily. The emergence of deep learning has also led to the renaissance of neural networks. The convolutional neural network (CNN) is one of the most popular types of artificial neural networks and is used widely for the classification of different types of images.

This paper presents a nine-layer deep CNN model for the classification of heartbeats into five different classes (non-ectopic, supraventricular ectopic, ventricular ectopic, fusion, and unknown beats) through the analysis of the patterns in electrocardiogram (ECG) signals. The proposed CNN model consists of three convolution layers, three max-pooling layers, and three fully connected layers. The CNN model was trained using the back-propagation technique with a sample size of 10. The ECG signals were taken from the open-source PhysioBank MIT-BIH Arrhythmia dataset. A total of 109,449 ECG beats were extracted for this study. This CNN model achieved accuracies of 89.07 and 89.3 percent in noisy and noise-free ECGs, even when trained with the highly imbalanced data (original dataset).

The proposed CNN architecture could also be used as a reference model for many other classification tasks (through the analysis of patterns found in other signals) in the field of medical science using the concept of transfer learning. Therefore, the paper is recommended for students and researchers exploring the applications of deep learning in medical science.