Heart rate variability feature selection method for automated prediction of sudden cardiac death

Sudden cardiac death is often caused by a major heart dysfunction with approximately 80% of occurrences attributed to ventricular arrhythmias. This paper proposes a novel ventricular arrhythmia prediction method to differentiate ventricular fibrillation and ventricular tachycardia from non-life-threatening cardiac activities, which may result in lower rates of unnecessary therapeutic intervention. Time series, frequency, bispectrum and nonlinear cardiac data can be analyzed on implantable devices such as implantable cardioverter defibrillators, to detect a life-threatening event. In order to reduce the computational overhead of the detection method, a feature selection technique is applied to reduce the number of features. Signal buffer lengths are examined, where one-minute and five-minute signal durations are processed. The features are then applied to classifiers and results from a selection of classification algorithms are compared. The proposed method achieves 88.8% sensitivity and 94.2% specificity using only six HRV features with a k-nearest neighbor classifier while processing a 5-minute window of R-R interval signals. Using a 1-minute window of R-R interval signals results in 86.9% sensitivity and 93.5% specificity, with a support vector machine classifier. The results show that the proposed method, using between four and eleven features, outperforms related work in the literature.