A cardioid based technique to identify Premature Ventricular Contractions

Premature Ventricular Contraction (PVC) can occur in healthy people of any age and is linked to mortality when associated with myocardial infarction. There has been voluminous research that focuses on PVC detection. However, pre-processing techniques such as wavelet filtering may cause delay by extracting features on frequency domain rather than time domain. Furthermore, many classifiers are only suitable for powerful processing systems. In this paper, we propose a new patient-specific classification technique to detect PVC by using two dimensional cardioid loops obtained from QRS complexes. Pre-processing time is reduced significantly since cardioid loops can be drawn directly from raw QRS complexes. The feature set comprises x-y coordinates of centroid, upper, lower, left and right extreme points of each cardioid loop. We conducted experiments over 20 subjects of the MIT/BIH arrhythmia database and obtained an average detection accuracy of 99.60%, average sensitivity of 97.43%, and average positive predictive value of 98.62%.