Improved detection and classification of arrhythmias in noise-corrupted electrocardiograms using contextual information within an expert system

Abstract

The authors are developing an expert-system electrocardiogram (ECG) arrhythmia detector (HOBBES) for automated, long-term rhythm analysis. HOBBES employs rules and procedures that emulate how human experts analyze ECGs. This paper describes methods that HOBBES employs for improving error detection and correction in processing noisy ECGs. During periods of clean data, HOBBES develops a knowledge base that describes typical beat shapes, typical interbeat intervals between beats of different types, and patterns of beat sequences that it has observed. During periods of noisy data, HOBBES applies the information learned from the clean data to reject artifact and classify beats. HOBBES was evaluated in a noise-stress test using 35 half-hour ECG records containing a mixture of supraventricular and ventricular ectopy in normal sinus rhythm. In comparison with a classical arrhythmia detector (ARISTOTLE), HOBBES increased the number of correctly classified beats and enhanced the rejection of artifact.