Predicting patterns of epicardial potentials during ventricular fibrillation.

Abstract

Ventricular fibrillation (VF) is a fatal cardiac arrhythmia, characterized by uncoordinated propagation of activation wavefronts in the ventricular myocardium. Short-term predictions of epicardial potential fields during VF in pigs were attempted using linear techniques, and prediction accuracy was measured at various stages during sustained episodes. VF was induced in five pigs via premature electrical stimulation. Unipolar electrograms were recorded from an epicardial array of 506 electrodes in a 22 x 23 array with 1-mm spacing. Optimal spatial basis functions (modes) and time-varying weighting coefficients were found using the Karhunen-Loeve decomposition. Linear autoregressive (AR) models incorporating the dynamics of only a few spatial modes led to predicted patterns that were qualitatively similar to observed patterns. Predictions were made 0.256 s into the future, based on 0.768 s of past data, over an area of approximately 5 cm2 on the ventricular epicardium. The mean squared error of predictions varied from as much as 1.23 to as little as 0.14, normalized to the variance of the actual data. Inconsistency in long-term forcasts is partly due to the limitations of linear AR models. Changes in predictability, however, were consistent. Predictability varied inversely with spatial complexity, as measured by the mean squared error of a five-mode approximation. Predictability also increased significantly during the first minute of VF.