Prediction of Ventricular Fibrillation Using Support Vector Machine

Abstract

Sudden cardiac death (SCD) remains one of the top causes of high mortality rate. Early prediction of ventricular fibrillation (VF), and hence SCD, can improve the survival chance of a patient by enabling earlier treatment. Heart rate variability analysis (HRV) has been widely adopted by the researchers in VF prediction. Different combinations of features from multiple domains were explored but the spectral analysis was performed without the required preprocessing or on a shorter segment as opposed to the standards of The European and North American Task force on HRV. Thus, our study aimed to develop a robust prediction algorithm by including only time domain and nonlinear features while maintaining the prediction resolution of one minute. Nine time domain features and seven nonlinear features were extracted and classified using support vector machine (SVM) of different kernels. High accuracy of 94.7% and sensitivity of 100% were achieved using extraction of only two HRV features and Gaussian kernel SVM without complicated preprocessing of HRV signals. This algorithm with high accuracy and low computational burden is beneficial for embedded system and real-time application which could help alert the individuals sooner and hence improving patient survival chance.