Patterns in Hilbert Transforms and Wigner-Ville Distributions of Electrocardiogram Data R.M.S.S. Abeysekera, R.J. Bolton", L.C. Westphal, B. Boashash University of Queensland, Brisbane, Australia. *University of Saskatchewan, Saskatchewan, Canada ABSTRACT The computation and characteristics of two new representations of single lead electrocardiogram data, both based upon the analytic signal corresponding to the waveform, are presented. The first method, employing a complex plane plot of the analytic signal, emphasizes waveform shape independently of time base: the second, a time--frequency representation using the Wigner-Ville distribution, appears to highlight instantaneous frequency variation within the waveform. Both are potentially useful when the patterns are displayed using contour plots, and a specific computerised pattern recognition technique is suggested and some results presented. INTRODUCTION Electrocardiograms (ECGS) are important tools in diagnosis of heart irregularities and hence have been extensively researched €or the possibility of computer assistance in the diagnostic process. This assistance is largely restricted to pattern recognition based primarily upon features extracted from the ECG waveform in a manner analogous to human cardiologists. Thus the major tools now available, such as multivariate analysis [based on Ref. 11 and AZTEC systems [2] depend on details oE the plots of voltage vs. time. Only limited use has been made of spectral information derived from the Fourier transform [e.g. 31. and other transforms have hardly been considered. Our research has been concerned with alternative representations of the ECG data. In this paper we review our results to date in applying the Hilbert Transform and the Wiqner-ViLle Distribution to such signals. These provide two different ways to provide additional emphasis to certain aspects of the waveforms and hence to allow alternative features of the heartbeats to be studied and used in the diagnostic process. In both of these methods, we first form the analytic signal z(t) associated with a single-lead ECG signal x(t) by calculating where H[ 1 = Hilbert transform operator. A typical ECG waveform and its Hilbert transform are shown in Figure 1. In our first method, discussed in the following section, z(t) is plotted in the complex plane with t as a parameter. In an alternative approach presented in Section 3, the Wigner-Wlle time-frequency distribution, the Fourier transform of an instantaneous auto-correllation estimate of the analytic signal, is computed. Because both result in contour plots, a potentially useful pattern recognition method is presented in Section 4. Section 5 presents some preliminary results and a brief summary. Al.1 heartbeats processed are taken from the MIT-BIH database [71. Typical heartbeats were about 280 samples, taken at 360 samples per second. The Hilbert transforms were performed using FIR filters. Coding was primarily in the C language, and calculations were performed on Digital Equipment Corporation PDP 11/40 and VAX 11/750 minicomputers. HILBERT TRANSFORM PLOTS When the analytic signal is plotted in the complex plane with time t as the parameter, the result (Figure 2) is very similar in general appearance to a standard vectorcardiogram (VCG) representation, although the detail differences are substantial and the underlying physical interpretation of the VCG is not applicable. In interpreting the analytic signal plot, one should observe that a. The basic motion is counter-clockwise. b. The largest loop corresponds to the QRS complex, the next to the T wave, and the smallest to the P wave. c. Unless there is a voltage offset in the original data (which yields a real axis oEfset), quiescent periods map into the origin of the polar plot. ICASSP 86, TOKYO 1793