Automatic Premature Ventricular Contraction detection in photoplethysmographic signals

The purpose of this study was the development and investigation of the automatic Premature Ventricular Contraction (PVC) detection and classification method using Photoplethysmographic (PPG) signals. The main issue of using PPG for arrhythmia detection are the artefacts which may be falsely detected as an arrhythmic pulses. The method is based on 6 PPG features, obtained in 12 s analysis frame. The features are peak-to-peak intervals and PPG power derived features. The fundamental frequency of the PPG was used for feature extraction and normalization. The Artificial Neural Network with back-propagation was used for the PPG pulse classification. The PPG signals from Physionet MIMIC II and MIMIC databases were used for algorithm training and testing. PPG were annotated by referring to synchronously registered ECG signals. The method was evaluated by calculating sensitivity and specificity which for the two main PVC types are 96,05 / 95,37 % and 99,85 / 99,80 %, respectively. The study results suggest that PPG can be used for the reliable PVC detection.