DETECTION OF VENTRICULAR-TACHYCARDIA USING SCANNING CORRELATION-ANALYSIS

Cross correlation is an accurate method for distinguishing normal sinus rhythm (NSR) from ventricular arrhythmias. The computational demands of the method, however, have prohibited development of an implantable device using correlation. In this study, temporal data compression prior to correlation analysis was used to reduce the total number of computations. Unipolar and bipolar intracardiac electrograms of NSR and 23 episodes of ventricular tachycardia (VT) from 23 patients were obtained from a right ventricular apex electrode catheter during routine electrophysiology studies. The data were filtered (1-11 Hz), digitized (250 samples/sec) and temporally compressed to 50 samples/sec. Data compression removed four out of every five samples by only saving the sample with the maximum excursion from the last saved sample. The average squared correlation coefficient (r2) was computed for the NSR and VT episodes using each patient's NSR waveform as a template. In all 23 patients, the r2 values showed large separation between NSR versus VT in both unipolar (0.93 +/- 0.05 vs 0.20 +/- 0.16, P < 0.005) and bipolar (0.91 +/- 0.07 vs 0.17 +/- 0.11, P < 0.005) electrode configurations using template lengths of 80% the intrinsic interval (avg +/- SD). Narrow templates (40% intrinsic interval or less) often resulted in multiple r2 peaks during each heart cycle and degraded the r2 separation (n = 10, P < 0.005). High pass filtering at 3 Hz also degraded the r2 separation (n = 10, P < 0.05). Standard noncompressed correlations indicated that data compression had negligible effects on the results. Thus, a computationally efficient cross correlation method was found to be a reliable detector of VT. The results suggest that the T wave, as well as the QRS complex, is useful in VT recognition.