Rule-based rough-refined two-step-procedure for real-time premature beat detection in single-lead ECG

Abstract

Objective: Premature beats (PB), typically presenting as premature ventricular contractions (PVC) and premature atrial contractions (PAC), may foreshadow stroke or sudden cardiac death. Approach: A rule-based real-time PB detection system was proposed for timely diagnosis of common PAC and PVC in an ambulatory setting and to reduce the cognitive load for physicians. The proposed method consists of three procedures: (1) extraction of the RR interval, QRS complex template, width and height; (2) rough detection of PB candidates using rules corresponding to abnormality in rhythm and morphology; and (3) refined detection using three types of correction. The method was trained using randomly selected single-lead waveforms sourced from the China Physiological Signal Challenge 2018 (CPSC2018) database, and the method was tested on the 12-lead CPSC2018 database, the MIT-BIH-AR database and the wearable ECG database. Main results: Four quantitative parameters, namely sensitivity, positive predictive value, accuracy and F1 measure, were used to assess performance. The F1 measure for normal beats, PACs, and PVCs were 99.37, 90.6, and 90.85% in training data (93.61% across all beats). Satisfactory results on the 12-lead CPSC2018 database indicated that the method had a good generalization ability between leads. Although the results on the MIT-BIH-AR database were not comparable with other methods, it showed stability in different testing databases. In addition, the test results on wearable ECGs manifested that the method was robust and could provide a good algorithm basis for IoT applications. Significance: We have developed a rule-based method for real-time PB detection in single-lead ECG, which balances the computational complexity and recognition accuracy, indicating the clinical significance of the method.