Optimizing the Detection of Characteristic Waves in ECG Based on Processing Methods Combinations

Abstract

Accurate detection of characteristic electrocardiogram (ECG) waves is necessary for ECG analysis and interpretation. In this paper, we distinguish four processing steps of detection algorithms: noise and artifacts reduction, transformations, fiducial marks selection of wave candidates, and decision rule. Processing steps combinations from several detection algorithms are used to find QRS, P, and T wave peaks. In addition, we consider the search window parameter modification based on waveform templates extracted by heart cycles clustering. The methods are extensively evaluated on two public ECG databases containing QRS, P, and T wave peaks annotations. We found that the combination of morphological mathematical filtering with Elgendi's algorithm works best for QRS detection on MIT-BIH Arrhythmia Database (detection error rate (DER = 0.48%, Lead I). The combination of modified Martinez's PT and wavelet transform (WT) methods gave the best results for P wave peaks detection on both databases, when both leads are considered (MIT-BIH arrhythmia database: DER = 32.13%, Lead I, DER = 42.52%, Lead II; QT Database: DER = 21.23%, Lead I, DER = 26.80%, Lead II). Waveform templates in combination with Martinez's WT obtained the best results for T wave peaks detection on QT database (DER = 25.15%, Lead II). This paper demonstrates that combining some of the best proposed methods in literature leads to improvements over the original methods for ECG waves detection while maintaining satisfactory computation times.

Keywords

• ECG
• characteristic waves
• automatic detection algorithms
• clustering
• expert system
• biomedical signal analysis