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

Area of Science:

• Cardiovascular Medicine and Signal Processing
• Biomedical Engineering

Background:

• Sudden cardiac death (SCD) affects millions globally, often without warning, posing a critical health challenge.
• Ventricular fibrillation is a key factor in SCD pathophysiology, with symptoms appearing shortly before the event.
• Timely prediction of SCD is essential for effective resuscitation and improving patient outcomes.

Purpose of the Study:

• To develop and evaluate a method for predicting sudden cardiac death (SCD) using time-frequency analysis of electrocardiogram (ECG) signals.
• To assess the efficacy of machine learning algorithms (SVM, KNN) combined with feature selection techniques (T-test, ANOVA) for SCD prediction.

Main Methods:

• Utilized two public ECG datasets: Sudden Cardiac Death Holter and MIT-BIH Normal Sinus Rhythm.
• Segmented the 60-minute interval preceding ventricular fibrillation into one-minute segments.
• Applied empirical mode decomposition (EMD) for time-frequency sub-band analysis, extracted nonlinear features, and employed SVM and KNN classifiers with T-test and ANOVA feature selection.

Main Results:

• The proposed method achieved high accuracy in predicting SCD within the 60 minutes preceding the event.
• ANOVA feature selection combined with SVM and KNN yielded the highest average accuracy rates: 93.51% (ANOVA-SVM) and 93% (ANOVA-KNN).
• T-test feature selection also demonstrated strong performance, with average accuracy rates of 93.29% (SVM) and 93.41% (KNN).

Conclusions:

• The developed approach using time-frequency analysis and machine learning shows significant promise for predicting sudden cardiac death (SCD).
• Feature selection techniques, particularly ANOVA, enhance the accuracy of SCD prediction models.
• This predictive capability could lead to improved early intervention strategies and better patient survival rates.