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

Accurate and efficient ECG signal classification plays a vital role in ensuring prompt diagnosis and effective management of cardiac conditions. This research proposes a novel hybrid framework that incorporates the Wavelet Packet Transform (WPT), a Modified Convolutional Neural Network (CNN), and a Multi-Branch Transformer to overcome challenges related to noise resilience, computational efficiency, and adaptability in ECG signal analysis. The model’s performance was validated using both the MIT-BIH database and a private clinical dataset, where it achieved remarkable accuracy rates of 99.48% and 99.17%, respectively, along with high recall scores in both noisy and clean environments. Leveraging WPT for multi-resolution analysis facilitated accurate feature extraction, while the hybrid architecture seamlessly integrated the CNN’s capability to detect localized signal features with the Transformer's proficiency in capturing long-range temporal relationships. Computational efficiency was enhanced through optimized pre-processing and the design of a lightweight transformer, achieving an average execution time (AET) of 2.25 seconds, significantly outperforming other approaches. Comparative evaluations against state-of-the-art models demonstrated the proposed method’s adaptability to diverse datasets and robustness against signal noise and variability. These results underscore the model’s potential for real-time, noise-robust ECG interpretation in clinical settings, addressing persistent limitations in existing approaches, such as susceptibility to noise, high computational demands, and restricted generalization capabilities.