Comparison of polynomial coefficients for pattern recognition of sinus rhythm (SR) and ventricular tachycardia (VT)

A new algorithm to distinguish normal sinus rhythm (SR) and ventricular tachycardia (VT) waveforms based on coefficients of a Legendre polynomial approximation has been developed. The application of this technique is intended for rhythm distinction of intraventricular signals such as those sensed by implantable cardioverter defibrillators (ICDs). In this study of 10 patients with both SR and VT, a 10th order polynomial fit produced a waveform discrimination success rate of 100% for 8 patients, while for the two remaining patients the success rates were 69% and 96%. The effect of using fewer coefficients was also analyzed using coefficient reduction techniques based on probability of error (POE), average correlation coefficient (ACC), and eigenvectors (EV). It was found that for SR, a better discrimination rate could be achieved by using 8 coefficients instead of 21 by the POE method. Similarly 15 coefficients of VT gave better discrimination rates.