An Efficient Method for ECG Signal Classification using the Integration of Wavelet Packet Transform and Multi-branch Transformer Model

Abstract

The electrocardiogram (ECG) serves as a critical tool in cardiac monitoring, offering insights into heart conditions. However, classical ECG classification approaches suffer from noise interference, calculation complexities, and adaptability across different databases. This article presents a hybrid deep learning algorithm that integrates Wavelet Packet Transform (WPT), an improved Convolutional Neural Network (CNN), and a Multi-branch Transformer model to eliminate these challenges. By using WPT for multi-resolution decomposition, the introduced technique efficiently extracts critical features while removing the noise. The improved CNN caught localized signal transitions, while the Multi-branch Transformer algorithm adaptively represents long-range temporal dependencies, obtaining accurate classification of heart diseases. Validation tests were carried out using the MIT-BIH Arrhythmia database and a private clinical database, obtaining accuracy rates of 99.48% and 99.17%, respectively. The algorithm’s calculation complexities were significantly reduced, producing an average execution time (AET) of 2.25 s, outstanding over state of the art. Comparative assessments against state-of-the-art methods underscore the generalizability, efficiency, and robustness of the suggested approach, making it suitable for clinical and real-time tasks.