A novel method for early prediction of sudden cardiac death through nonlinear feature extraction from ECG signals.

Sudden cardiac death (SCD) is a critical cardiovascular issue affecting approximately 3 million individuals globally each year, often occurring without prior noticeable symptoms. While the precise etiology of SCD remains elusive, ventricular fibrillation is believed to play a pivotal role in its pathophysiology. Given that symptoms typically manifest only an hour before the event, timely prediction is crucial for effective cardiac resuscitation. This study aims to predict SCD using time-frequency analysis of ECG signals. We utilized two online datasets: the Sudden Cardiac Death Holter dataset and the MIT-BIH Normal Sinus Rhythm dataset. Our proposed method involves segmenting the 60-min interval preceding ventricular fibrillation into one-minute segments, which are then decomposed into time-frequency sub-bands using empirical mode decomposition (EMD). Nonlinear features are extracted from these decomposed signals, followed by classification using support vector machines (SVM) and K-nearest neighbors (KNN) algorithms. To enhance classification accuracy, we employed two statistical feature selection techniques: T-test and ANOVA. Results indicate that using the ANOVA feature selection method in conjunction with SVM and KNN algorithms achieves high accuracy in predicting SCD. Specifically, the average accuracy rates for the 60 min preceding SCD were 93.51% for ANOVA-SVM and 93% for ANOVA-KNN. With T-test feature selection, the average accuracy rates were 93.29% for SVM and 93.41% for KNN. These findings demonstrate the promising performance of our proposed approach in predicting SCD, potentially contributing to improved early intervention strategies and patient outcomes.