Detecting premature ventricular contractions in ECG signals with Gaussian processes

The aim of this work is twofold. First, we propose to investigate the capabilities of a new Bayesian approach for detecting premature ventricular contractions (PVCs), namely the Gaussian process (GP) approach. Second, we report an experimental comparison of different kinds of ECG signal representations, which are the standard temporal signal morphology, the discrete wavelet transform domain, the S-transform characteristics and the high-order statistics. In general, the obtained classification results show that the GP detector can compete seriously with state-of-the-art methods since it allows to yield better overall accuracy as well as better sensitivity. In addition, among the different kinds of features explored, those based on high-order statistics appear to be the best compromise between accuracy and computational time for PVC detection.