Accurate classification of premature beats, particularly premature atrial contractions (PACs) and premature ventricular contractions (PVCs), is clinically important but has not yet been thoroughly explored in PPG signals. While ECG remains the gold standard, PPG offers a non-invasive, user-friendly, and easy-to-use alternative that can be integrated into wearable devices, enabling continuous monitoring, early screening, and assessment across a broader population. Most existing studies focus on detecting only PVCs or classifying PPG segments rather than individual beats, and are often trained on small datasets, which limits their generalizability.
We propose a simple and effortless method based on only two inter-beat interval (IBI)-derived features to distinguish PACs from PVCs at the individual beat level in PPG. To overcome the limitation of small PPG datasets, the method is trained on simulated PPG signals generated from RR intervals derived from real ECG recordings of over 500 individuals, providing a large-scale and diverse dataset that enables a robust and generalizable approach.
The approach is evaluated on both simulated and real-world PPG recordings, achieving F1-scores of 0.77 and 0.76, respectively. These results demonstrate that the method is robust, computationally efficient, and capable of classifying PACs and PVCs at the beat level, making it suitable for practical applications in wearable PPG monitoring and screening across a broad population.