A comparison of human experts and computer algorithms in detecting and classifying beats in noise-corrupted electrocardiograms

Summarized is research that evaluated the performance of both human experts and automated arrhythmia detectors (ARISTOTLE and HOBBES) in processing noisy electrocardiograms (ECGs). Two studies were performed. The first study consisted of adding electrode motion artifact to clean ECG, and the second study consisted of adding high frequency (muscle noise) in addition to electrode motion artifact. A total of ten different 30-minute ECG records, each containing a mixture of normal beats, supraventricular beats and premature ventricular beats, were used for the first study. The second study used four of the ten tapes from the first study, but with additional noise. The results from both of the studies show that although HOBBES improved the performance of ARISTOTLE in processing noise ECGs, human experts consistently performed better in detecting and classifying beats because they have the ability to recognize beat shapes using distinct portions of QRS complexes