Unsupervised Classification of Premature Ventricular Contractions Based on RR Interval and Heartbeat Morphology

Abstract

Accurate automated detection of premature ventricular contractions from electrocardiogram requires a training set or expert intervention. We propose a fully automated unsupervised detection method. The algorithm first clusters morphologically similar heartbeats and then performs classification based on RR intervals and morphology. Tests on clinically recorded datasets show sensitivity of 94.7%, specificity of 99.6% and accuracy of 99.5%.