T WAVE ALTERNANS DETECTION USING SIGNAL PROCESSING AND MACHINE LEARNING TECHNIQUES

Abstract

Despite persistent challenges in accurately identifying high-risk patients, sudden cardiac death (SCD) remains a pressing concern. This study addresses the noninvasive detection of T-wave alternans (TWA), an indicator of cardiac abnormalities, through a comprehensive investigation of physiological foundations and traditional techniques. By developing convolutional neural networks (CNNs), this research pioneers the application of CNNs in accurately recognizing TWA in electrocardiograms (ECGs).

This non-invasive and proactive approach holds promise for reducing the incidence of SCD. To achieve these goals, meticulous preprocessing techniques for ECG signals from six patients, described in the previous study, "Machine Learning approach for TWA detection relying on ensemble data design", are performed. The preprocessing steps include baseline wander elimination, signal segmentation, windowing, alignment, background subtraction, and labeling. Furthermore, the rigorous preprocessing methods ensure the reliability and robustness of the CNN models, enhancing the overall accuracy of TWA detection. The study's computational efficiency enables real-time implementation in clinical practice, with a training process time of less than one hour and thirty minutes.

Filling a significant gap in the literature, this study provides an essential foundation for future investigations in TWA detection. The CNN models showcase notable accuracy (74.36%) and statistically significant precision (79.43%), indicating their potential to reliably detect TWA in ECGs.

These achievements highlight the potential impact of this research on patient care and the prevention of SCD. The findings contribute to a deeper understanding of TWA detection, opening avenues for further advancements in this field. By leveraging CNNs, future research can build upon this study to develop more accurate and effective methods for identifying high-risk patients and improving patient outcomes.