The Investigation and Validation of the \(\alpha\)-Stable Distribution Characteristics for Noises that Corrupt ECG Signals

The diagnostic accuracy and reliability of an unsupervised electrocardiogram (ECG) analysis system entirely depend on the response of ECG preprocessing stage. Unfortunately, ECG signal analysis faces the challenge of getting distorted by various noises and artifacts (physiological and non-physiological origin). Thus, designing a denoising technique capable of dealing with different noises in real time is challenging, so selecting a noise analysis model is particularly important. Based on the extensive survey of state-of-the-art techniques, it is noticed that all the denoising techniques are designed with an implicit assumption that noises distorting ECG signals are of Gaussian nature and therefore, are based on the Gaussian distribution noise analysis model. However, in practical scenarios, noises may not always have a Gaussian nature. Therefore, this paper puts forward a non-Gaussian \(\alpha\) $\alpha$ -stable distribution model for noise analysis from the perspective of ECG signal analysis. This distribution model encompasses Gaussian distribution as a special case. From rigorous simulations and analytical studies, this research offers statistical proof that the \(\alpha\) $\alpha$ -stable distribution noise model may effectively capture background noises corrupting ECG signals. Ultimately, the effectiveness of R-peak detection techniques and deep learning models is evaluated in the presence of two types of noise: Gaussian distribution and \(\alpha\) $\alpha$ -stable distribution. Finally, through intensive simulation studies, it is discovered that relying on the assumption of Gaussian background noise can be misleading when the actual noise follows a non-Gaussian \(\alpha\) $\alpha$ -stable nature.