ST-T complex of electrocardiogram (ECG) signal reflects an important phase of the ventricular repolarization. Many efforts have been made towards the ST-T analysis because of its clinical significance in diagnosing myocardial ischemia, infarction, etc. However, the identification of the ST segment fiducial points has not been standardized yet. The most often used time domain based methods are mainly empirically dependent and noise sensitive. In this research, ST-T complex automatic analysis including fiducial point detection and change representation of the ECG signals was investigated in the wavelet transform domain. Before ST segment analysis, it was particularly important to remove the baseline wandering for precise measurement of the ST segment. Due to the partial frequency overlap between the ST segment and the baseline wandering, the normally used digital filters cannot remove the overlapped frequency components without distorting the useful signal. Therefore, a wavelet adaptive filter structure was used for this purpose. Then, taking advantages of the multiple resolution ability of the wavelet transform, by analyzing the corresponding relations between the singular points of the ST segment and the wavelet transforms at different scales, we developed the identification rules at four different wavelet decomposition scales or frequency bands to identify the fiducial points of the ST segment. The proposed methods were tested using the standard MIT/BIH ECG database. The fiducial points identification results were compared with those manually annotated by the cardiologists from the Affiliated Hospital of Auhui Medical University, Hefei, China. This comparison showed a very good matching, which suggested the reliability of the proposed method. It is the case that some factors such as sport exercise, drugs, etc. may cause ischemia of the myocardial cells, which will change the morphological features of the ST segment and influence the singularity of the fiducial points. Based on this, we also investigated the relations between the ST segment change in time domain and the tendency of the local extrema change in different scales or frequency bands, and proposed a new approach to describe the ST segment change representations. Examples of ST segment change, such as ST elevation and ST depression were given.
© Copyright 2002 International Journal of Computer Science in Sport. Sciendo. Alle Rechte vorbehalten.