The Accuracy on the Common Pan-Tompkins Based QRS Detection Methods Through Low-Quality Electrocardiogram Database

摘要

QRS complex detection has been extensively studied over the past decades. Pan-Tompkins based methods were the most widely used QRS complex detectors. These methods could obtain high accuracy on high-quality clinical ECG data. However, the accuracy of the Pan-Tompkins based QRS detection methods on a low-quality ECG database should be clarified. In this paper, three Pan-Tompkins based QRS complex detectors were tested on both high and low-quality ECG databases, i.e., the original Pan-Tompkins method and the following improved versions of Pan-Tompkins based method with mean and median peak level estimation respectively. Two hundred 10-minute ECG recordings from the 2014 PhysioNet/CinC Challenge were used (100 from the high-quality database and 100 from the low-quality database). Accuracy was used as the performance criterion for the methods. All three Pan-Tompkins based QRS detection algorithms had high detection accuracies on the high-quality ECG database (> 99%), whereas relatively low detection accuracies were reported for the lowquality ECG database (76.03% was the highest result for the Pan-Tompkins based method with median peak level estimation and 74.49% was the lowest result for the original Pan-Tompkins method). We found that it is better to use the median peak level estimation to avoid the sudden amplitude change effects. In addition, the original Pan-Tompkins method had the highest computational efficiency. In this study, the performance of three common Pan-Tompkins based QRS complex detection algorithms were systematically analyzed and their noise responses were also tested. The conclusion generated by this study could potentially offer important reference for the reasonable use of those methods.