Arrhythmia Classification and Analysis on ECG Using Convolutional Networks and Two-fold Focal Loss

Abstract

Throughout recorded history, cardiovascular diseases have posed a persistent threat, claiming numerous lives. Effective and timely testing is pivotal in preventing fatalities. Among the available testing options, the Electrocardiogram (ECG) stands out as both practical and cost-effective, capable of diagnosing various abnormalities. Recently, there has been a notable emphasis on accurately classifying heartbeats. Traditionally, heartbeat analysis has been approached through manual or automated methods. Manual analysis involves cardiologists, while automated analysis relies on computational algorithms. Automated techniques have gained significant popularity in recent years and have achieved considerable success. However, despite this progress, there is still a need for further improvement to achieve deployable accuracy. Many current studies utilize deep learning models in a transfer learning approach for heartbeat classification. While transfer learning offers advantages, it also presents disadvantages such as domain mismatch, task-specific features, interpretability concerns, model bias, and generalization issues. Therefore, in this study, instead of employing transfer learning, a deep convolutional neural network combined with twofold focal loss is utilized for heartbeat classification. The proposed approach has demonstrated the ability to accurately classify five distinct arrhythmias according to the AAMI EC57 standard. Testing was conducted using the MIT-BIH and PTB Diagnostics datasets from PhysionNet. The results indicate that the proposed method achieves an average accuracy of 99.8% in classifying arrhythmias.