Premature Ventricular Contraction Detection Algorithm Based on Robust Feature Extraction

Abstract

An accurate and efficient premature ventricular contraction (PVC) detection algorithm for electrocardiography (ECG) signals was developed. To detect PVC accurately, the features should not only be chosen properly but also be determined precisely. Therefore, the following ways are adopted to improve the performance of PCV detection. Since many features for PVC detection are related to the amplitudes and locations of P, Q, R, S, and T points, we apply inverse-gradient weight functions to extract the baseline precisely and determine their amplitudes more accurately. Second, a location estimation mechanism is adopted to determine their locations precisely. Moreover, instead of PR intervals, QT intervals, and ST segments, which are hard to measure precisely, the distances among Q, R, and S points are adopted. Furthermore, the existences of P and T waves, the local and sub-global RR interval ratios, and a product-form score function are also applied for PVC determination. Simulations for the MIT-BIH Arrhythmia dataset show that the proposed algorithm achieves a sensitivity of 98.079% and a specificity of 99.306%. The proposed algorithm applies only 8 features but can achieve a very good performance.