A novel algorithm for the prediction and detection of ventricular arrhythmia

Abstract

The electrocardiogram signal (ECG) has been widely used as a diagnostic tool for many heart related diseases. In this work, we present a novel method for the prediction and detection of ventricular arrhythmia using a unique set of ECG features extracted from two consecutive cardiac cycles. Two databases of the heart signal recordings from MIT Physionet and the American Heart Association were used as training, test, and validation sets to evaluate the performance of the proposed method. Linear Discriminant Analysis was used to differentiate between normal and abnormal ECG signals. The discriminatory properties of the ECG features were evaluated by many k-fold cross validations. The proposed method achieved: an accuracy of 99.1%, sensitivity of 98.95%, precision of 98.39%, and area under the receiver operating characteristics curve of 99.97% on the out-of-sample validation data by tenfold cross validation with 5 s window size. Furthermore, the method was capable of predicting the onset of ventricular arrhythmia up to 3-hours earlier; this is the earliest prediction interval compared to any published work so far.