Automatic classification of arrhythmic heartbeats using the linear prediction model

This study developed an automatic heartbeat classification system based on the morphological features extracted using the first-order linear prediction model with two optimal filter coefficients and the RR interval features normalized by the heart rate of individual patient to reduce the effects of inconsistent heart rates among patients. Three heartbeat classes, normal beats, supraventricular ectopic beats and ventricular ectopic beats obtained from the MIT-BIH Arrhythmia Database, were used to test the performance of the proposed method. The ECG data were divided into training and testing datasets, each containing about 50,000 heartbeats. The training dataset was first used to establish the optimal linear discriminant classifier, and then the testing dataset was applied to evaluate the classification performance. The study results demonstrate that the sensitivity and positive predictive value of the proposed method were 88.7% and 99.4% for normal beats, 79.5% and 30.1% for supraventricular ectopic beats, and 88.6% and 57.7% for ventricular ectopic beats, respectively. If the RR interval features without normalization were used, the sensitivity and positive predictive value for supraventricular ectopic beats decreased to 62.5% and 24.0%, respectively.