Session S34.5

Methods for Initialization of Activation-Based Inverse Electrocardiography Using Graphs Derived from Heart Surface Geometry

B Erem*, PM van Dam, A Keely, JG Stinstra,
TF Oostendorp, DH Brooks

Northeastern University
Boston, MA, USA

The activation-based inverse problem of electrocardiography is non-linear in the desired activation times. Current solutions rely on iterative algorithms and are typically highly dependent on the initialization used. Thus there has been considerable interest in improved initialization approaches. Recent efforts include the critical point method of Huiskamp and Greensite and the shortest path - fastest route methods of van Dam, Oostendorp, and van Oosterom. The first of these, in the implementation commonly in use, projects single-node activations onto the noise subspace of the data. Small projections indicate nodes which "initiate" local activation spread, along with the associated activation timing. In effect it assumes that the final activation sequence is a superposition of independent single-node activations. The second approach uses the heart geometry to derive likely activation patterns and then compares the body surface potentials predicted by each such pattern to measured data using a correlation approach. Multiple such patterns are combined to create more complex activation sequences using a simple "first-arrived, first-chosen" approach. In this work we analyze the relationship between these two methods. We also suggest an alternative to the shortest path approach that may have computational advantages. We use a geometrically-derived graph adjacency matrix, as an alternative to the shortest path search, to represent the set of likely activation patterns to be tested. Each row of that matrix is then taken as a candidate "shortest path" type pattern, and the correlation to the resulting predicted body surface potentials is used to choose the best pattern. We also explore a possible combination approach, using the projection measure on selected subintervals of the data to find multiple likely activation patterns. We use epicardially stimulated data and associated geometries, recorded at the Cardiovascular Research and Training Institute in Utah, and body surface data and geometries supplied along with the ECGSim software, to compare results for a variety of activation patterns, as part of an effort to provide a common platform for comparison of methods for inverse electrocardiography.

(Abstract Control Number: 111)