Model-based detection and classification of premature contractions from photoplethysmography signals

The detection of arrhythmias from wearable devices is still an open challenge, while the availability of screening tools for the large population would allow reduced complications and costs. We propose a model-based approach to the detection and classification of premature contractions into atrial and ventricular. The extracted signal morphology and the deviations from the expected stationarity are used to detect and classify premature contractions. Our approach is self-contained, patient-specific and robust to mis-segmentation. Both model fit, and detection and classification accuracy of the proposed methods are evaluated on two real cases and a simulated dataset, and show promising results.