Multi-criteria Bayesian optimization of Empirical Mode Decomposition and hybrid filters fusion for enhanced ECG signal denoising and classification: Cardiac arrhythmia and myocardial infarction cases.

Abstract

This paper introduces a new advanced model for denoising and classification of ECG signals, focusing on the use of a hybrid filter and Bayesian optimization. The hybrid filter synergistically combines enhanced empirical mode decomposition (EEMD), Chebyshev Type II filters, Butterworth, Daubechies Wavelet, and Savitzky-Golay filters, leveraging their respective advantages for effective noise reduction while preserving the essential features of the ECG signal. We employ a multi-criteria Bayesian optimization process, using cross-correlation and mean squared error (MSE) as key metrics, to refine the filter parameters to further improve the signal quality. Additionally, we deploy a residual and parallel deep learning architecture adapted for the precise classification of cardiac conditions, such as arrhythmia and myocardial infarction. The results show a significant improvement in classification accuracy, from 96.63%-97.86% to 97.24%-98.62% after filtering, and reaching 98.03%-99.61% after optimization for arrhythmia. For myocardial infarction, accuracy increases from 96.29%-97.42% to 97.34%-98.44% after filtering, and from 98.13%-99.07% after optimization. This holistic and multi-criteria approach enhances the reliability of medical diagnostics and represents a significant advancement in the processing of biomedical signals.