The prediction of ventricular fibrillation based upon HRV signal using combination of genetic algorithm and neural networks

Abstract

Nowadays ventricular fibrillation has become one of the most serious cardiac rhythm disturbance in the world. It is worth predicting this problem and also achieving valuable chance for clinical therapy. At the time of happening VF some symptoms in heart rate variability (HRV) signal are apparent, which can be used to predict VF. Before happening VF, finding these symptoms are cumbersome and this problem leads to the fact that the proposed approach becomes weaker because it is desirable to reduce the time of prediction. In this study, an algorithm with the purpose of predicting VF is presented, which is based on extracting linear, frequency and non-linear features from HRV signal. Extracted features are validated through the use of t-test, and useful features are extracted by means of genetic algorithm (GA) and neural networks. Target function of GA is selected by considering the linear combination with positive rate of accuracy, sensitivity, specialty and negative rate in number of features. The most significant achievement of this study is taking no notice of frequency features of HRV signal, which represents the stress level and also relaxation of patient before happening VF. Moreover, the obtained results of this study are compared based on k-fold test and has achieved the accuracy of 96.67% in accordance with the non-linear features.