Detecting ECG characteristic points by novel hybrid wavelet transforms: an evaluation of clinical SCP-ECG database

The morphologic analyses of ECG waveforms have been used often by physicians. A robust ECG delineator that can identify the characteristic points under different kinds of diseases with various noise interferences is crucial in clinical ECG waveform analyses. In this study, hybrid wavelet transforms (WT) were used to develop a tool for ECG delineation. The decomposed detail levels of d4 and d3 based on the Haar discrete WT were used to locate the QRS complex. After finding the R wave, the QRS duration, the onset- and end-points of P and T waves were identified by a multilevel Haar wavelet packet analysis in a moving search window. Approximately 10000 beats based on four categories of 12-lead ECG records including normal (N), acute myocardial ischemia (AMI), hyperkalemia (HK), and atrial fibrillation (AF) were randomly selected from the previous established SCP-ECG database to evaluate the algorithms on detecting QRS complex and delineation. The results showed (1) the sensitivities of QRS complex detection were 100%, 99.51%, 99.72%, and 99.65% in N, AMI, HK, and AF ECG records with various noise interferences respectively; (2) the positive predictivities of QRS complex detection were 100%, 99.46%, 99.66%, and 99.88% in the previous 4 categories of ECG records; and (3) the onset- and end-points of P, T waves can also be detected based on wavelet packet analyses when signals were interfered by noise and baseline wandering. The algorithms developed in this study can be applied directly onto clinical 12-lead ECG records for waveform analyses with their high accuracy on characteristic point detection in various leads and diseases. They can also be applied onto Holter ECG systems for QRS detection because of their robust ability for single-lead detection