Clinical Sudden Cardiac Death Risk Prediction: A Grid Search Support Vector Machine Multimodel Base on Ventricular Fibrillation Visualization Features

• Main Authors: Xie, Chao Xin, Wang, Liang Hung, Yu, Yan Ting, Ding, Lin Juan, Yang, Tao, Kuo, I. Chun, Wang, Xin Kang, Gao, Jie, Abu, Patricia Angela R
• Format: text
• Published: Archīum Ateneo 2025
• Subjects: ECG; Sudden cardiac death prediction; SVM; Visualizing feature; Analytical, Diagnostic and Therapeutic Techniques and Equipment; Biomedical; Computer Engineering; Computer Sciences; Electrical and Computer Engineering; Medicine and Health Sciences; Physical Sciences and Mathematics
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/420; https://doi.org/10.1016/j.compeleceng.2024.110022
• Institution: Ateneo De Manila University

Sudden cardiac death (SCD) occurs when an individual experiences ventricular fibrillation (VF) and does not receive intervention within several minutes. Predicting SCD or VF can provide medical professionals with additional time to perform rescues, thereby reducing mortality. This study proposes a novel high-efficiency grid search-based support vector machine algorithm (GS-SVM) for SCD risk prediction. It significantly reduces the time required to construct models. Nineteen VF-related visualization features (i.e., mean, standard deviation, approximate entropy of RR interval, QRS duration, corrected QT interval, Tp-Te interval, Tp-Te/QT ratio, and T-wave amplitude, as well as heart rate variability) were innovatively extracted from electrocardiogram (ECG) signals. Next, a distribution analysis of the features was conducted to convincingly highlight the differences between those derived from SCD samples and healthy controls. Furthermore, the GS-SVM algorithm was used to construct five SCD risk prediction models in accordance with the interval before the occurrence of VF. The highest accuracy of 95.78 % was obtained for predicting VF when 30 min before its occurrence. In addition, this study extended the prediction time to 70 min and achieved an accuracy of 90.08 %. Finally, to demonstrate the generalizability and clinical applicability of the proposed algorithm, two external datasets were used, the Creighton University Ventricular Tachyarrhythmia Database and the clinical Fujian Provincial Hospital Database. The overall accuracies achieved on them are 83.12 % and 93.75 %, respectively. The proposed algorithm effectively predicts the SCD at an earlier stage. Additionally, it can be integrated into ECG monitoring systems to provide real-time alerts for individuals.