Improved arrhythmia detection in noisy ECGs through the use of expert systems

The authors present a method of improving error detection and correction in noise. They characterize a signal in clean data and interpret noisy data by generating and evaluating all plausible transmitted signals. They develop the technique in the context of real-time arrhythmia analysis as a part of the expert system CALVIN (W.K. Muldrow et al., Comput. in Cardiology, p.21-6, 1986). By learning timing intervals between normal and ventricular beats in clean data, CALVIN can generate and evaluate hypotheses of sequences of beats to explain noisy data. CALVIN is evaluated in a noise-stress test using series 1000 through 5000 of the AHA Arrhythmia Database. The results indicate that the CALVIN-aided system increases the number of beats properly classified and greatly enhances rejection artifact