Sudden Cardiac Death Prediction by Fusing Electrocardiogram and Heart Rate Variability Signals

The prediction of a Sudden Cardiac Death (SCD) long enough before its occurrence is vital for cases outside the hospital. This study investigate the effect of the simultaneous application of Electrocardiogram (ECG) and Heart Rate Variability (HRV) signals in the SCD prediction 60 minutes before its incidence. To do so, first, the SCD prediction was performed in each of the one-minute intervals by different groups of linear and nonlinear ECG and HRV features using the Support Vector Machine (SVM) classifier. The results showed that the best accuracy for SCD prediction was 91.23%. Next, all features were ranked locally in each of the one-minute intervals before the incidence of the death using the Minimum Redundancy and Maximum Relevancy (MRMR) method. Then, the SCD was predicted by applying four top local features from the ECG and HRV signals in each one-minute interval an hour before the death, showing a mean accuracy and sensitivity of 99% and 98.76%, respectively. Finally, by selecting the four most effective features according to the number of times they have been chosen in all one-minute intervals, the mean accuracy and sensitivity of SCD prediction were calculated at 96.15% and 95.07%, respectively. Additionally, since there is a similarity between the ECG signal of the pre-SCD and the Congestive Heart Failure (CHF), the classification of the Normal, CHF, and pre-SCD was performed, indicating a mean accuracy of 79.7%; it was also discovered that the Normal data could be separated from the SCD and CHF data with higher accuracy.