Evaluation of patient electrocardiogram datasets using signal quality indexing

Abstract

Electrocardiogram (ECG) is widely used in the hospital emergency rooms for detecting vital signs, such as heart rate variability and respiratory rate. However, the quality of the ECGs is inconsistent. ECG signals lose information because of noise resulting from motion artifacts. To obtain an accurate information from ECG, signal quality indexing (SQI) is used where acceptable thresholds are set in order to select or eliminate the signals for the subsequent information extraction process. A good evaluation of SQI depends on the R-peak detection quality. Nevertheless, most R-peak detectors in the literature are prone to noise. This paper assessed and compared five peak detectors from different resources. The two best peak detectors were further tested using MIT-BIH arrhythmia database and then used for SQI evaluation. These peak detectors robustly detected the R-peak for signals that include noise. Finally, the overall SQI of three patient datasets, namely, Fantasia, CapnoBase, and MIMIC-II, was conducted by providing the interquartile range (IQR) and median SQI of the signals as the outputs. The evaluation results revealed that the R-peak detectors developed by Clifford and Behar showed accuracies of 98% and 97%, respectively. By introducing SQI and choosing only high-quality ECG signals, more accurate vital sign information will be achieved.