Spectral analysis of electrograms during ventricular tachycardia in a canine model: relation with epicardial isochronal maps.

The purpose of this study was to assess the capability of magnitude-squared coherence and bicoherence to differentiate monomorphic ventricular tachycardia (MVT) and polymorphic ventricular tachycardia (PVT) in a canine model and to relate these results to the epicardial isochronal maps on a beat-to-beat basis. Unipolar electrograms were simultaneously recorded from the surface of both ventricles with a 127-lead sock electrode array in 12 open-chest anesthetized dogs. The sampling frequency was 500 Hz. Atrioventricular block was induced by formaldehyde injection into the atrioventricular node. The left anterior descending coronary artery was occluded for 60 minutes under ventricular pacing (140 beats/min). During reperfusion, 12 MVT episodes lasting more than 42 seconds were recorded. Left stellate ganglion stimulation induced five PVT episodes lasting more than 42 seconds. Each of these recordings was divided into seven segments of 3,072 points (6.144 seconds). After visual selection, 104 segments were extracted and classified as 73 MVT and 31 PVT segments. Magnitude-squared coherence was estimated as the cross-spectrum from two epicardial signals (on the right and left ventricles, respectively), normalized with the respective autopower spectrum. Bicoherence was estimated as the bispectrum normalized with the autopower spectrum. Magnitude-squared coherence correctly identified 96% of MVT and 81% of PVT segments for a total accuracy of 91%. Bicoherence estimated with the left ventricular lead correctly identified 100% of MVT and 77% of PVT segments with an accuracy of 93%. Beat-to-beat epicardial maps of MVT displayed a cluster of sites of origin close to the reperfusion area, while the sites of origin from beats during PVT were much more dispersed over both ventricles. A strong and significant correlation was found between the number of electrodes with the earliest epicardial activation and coherence (r = .76, P < .0001) and bicoherence (r = .68, P < .0001), respectively. A high and significant correlation was also found between both spectral estimators (r = .74, P < .0001). Coherence and bicoherence discriminated accurately between MVT and PVT. Coherence achieved better results compared with bicoherence. Coherence and bicoherence measurements showed a quantitative relation with the spatial dispersion of the sites of origin. Both spectral techniques seemed powerful enough to be used in the development of implantable devices.