A Clinically interpretable rule-based algorithm for detection of premature ventricular contractions

Abstract

Premature Ventricular Contractions (PVCs) are a high-risk arrhythmia in patients with pre-existing cardiac conditions, often requiring long-term electrocardiogram (ECG) monitoring for timely detection and appropriate medical intervention. Currently, both rule-based algorithms and machine learning (ML) methods are actively being researched for automated PVC detection. Rule-based methods are considered older techniques with lower accuracy and poor robustness to noise, while ML algorithms are known for higher detection accuracy. However, despite their performance, ML models often suffer from limited generalizability and poor interpretability, which restrict their clinical adoption. Hence, in this work, we address the limitations of traditional rule-based algorithms—specifically their lower accuracy and noise robustness—by proposing a novel rule-based algorithm with significantly improved performance. The new algorithm emulates the diagnostic logic of cardiologists by leveraging clinically meaningful features and decision rules. Evaluated on the unseen INCART database, the algorithm achieves a high F1-score of up to 91.20%, demonstrating strong accuracy and generalizability, while offering a practical and interpretable alternative to ML-based approaches.