An Optimal Methodology for Early Prediction of Sudden Cardiac Death Using Advanced Heart Rate Variability Features of ECG Signal

• Published in: Arabian journal for science and engineering (2011) 2024-05, Vol.49 (5), p.6725-6741
• Main Authors: Karimulla, Shaik, Patra, Dipti
• Format: Article
• Language: English
• Subjects: Algorithms; Cardiovascular disease; Coronary artery disease; Death; Engineering; Gabor transformation; Heart rate; Humanities and Social Sciences; Image enhancement; Lifesaving; multidisciplinary; Research Article-Electrical Engineering; Science; Signal classification
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-023-08457-6

Sudden cardiac death (SCD) is a common cause of mortality in countries all over the world. A person will lose consciousness within a few minutes of exposure to SCD and ultimately die. Due to the limitations of SCD prediction algorithms that are currently available provide poor efficiency with minimal prediction time. Thus, developing a reliable and precise system is essential for accurate early-stage SCD prediction much time prior to its onset, which will save many people’s lives worldwide. It is essential to develop accurate methods for classifying ECG signals of various cardiac diseases that cause for the development of SCD. The present study proposes an optimal methodology for early prediction of SCD 1 h prior to its onset by extracting features from heart rate variability (HRV) signals of normal and abnormal conditions, specifically Normal sinus rhythm (NSR), Sudden cardiac death (SCD), coronary artery disease (CAD), congestive heart failure (CHF), and atrial fibrillation (AF). The HRV signal derived from the ECG signal is artifact-corrected, and 77 advanced features were extracted using the time domain, frequency domain, nonlinear methods, and Constant-Q Non-Stationary Gabor Transform (CQ-NSGT)-based image features. The proposed method using advanced extracted features yielded an enhanced prediction accuracy of 96.43%, with the optimal feature subset derived from the two-stage feature selection algorithm ANOVA-Sequential forward selection (SFS) followed by the Gradient boosting classification algorithm. The experimental outcomes of the proposed method and the exhaustive result analysis show that the proposed method has viable potential for early prediction of SCD before 1 h.