Electrocardiogram (ECG) Augmentation and Classification Using Deep Learning Techniques

Abstract

Electrocardiograms (ECGs) play a vital role in the clinical diagnosis of heart diseases. An ECG record can be used to detect the abnormalities of the heart and to discover numerous arrhythmias. The screening of cardiac arrhythmias requires a detailed study of the ECG records by the cardiologists, this process is time-consuming and too difficult. Hence, the automation process of arrhythmias identification and ECG analysis is crucial in the medical field. Medical datasets like the MIT-BIH arrhythmia dataset are often very limited and strongly imbalanced, which makes training the models— especially deep learning models—technically challenging, and the models will tend to be biased in favor of classes that contain a large number of samples. This thesis proposes a novel data-augmentation technique using generative adversarial networks (GANs) to restore the balance of the dataset and improve the classification of ECG heartbeats. Furthermore, two deep learning approaches—an end-to-end approach and a two-stage hierarchical approach— in addition to multiple deep convolutional neural networks (CNNs) and Recurrent Neural Networks (RNNs) are proposed to eliminate hand-engineering features by combining feature extraction, feature reduction, and classification into a single learning method.