Improving Localizing Cardiac Geometry Using ECGI

Jake Bergquist1, Jaume Coll-Font2, Brian Zenger1, Lindsay Rupp1, Wilson Good3, Dana Brooks4, Rob MacLeod1
1University of Utah, 2Boston Children's Hospital, 3The SCI Institute, 4Northeastern University


Introduction: Electrocardiographic Imaging (ECGI) is a promising tool for noninvasive diagnosis of cardiac dysfunction using body surface potentials (BSPs). ECGI requires a model of the torso, and one source of error is inaccuracy in the position of the heart, for example from respiratory movement. In previous work we presented a method to localize the heart from BSPs when heart surface potentials are known. Here we extend this method to ECGI, when the heart potentials are not known.

Methods: We parameterize the position of the heart with 6 degrees of freedom. We assume that consecutive heartbeats have the same cardiac source and the primary source of ECGI error is cardiac localization. We use an iterative coordinate descent optimization. In each iteration, a single inverse solution is computed with current per-beat estimates of cardiac location using BSPs from multiple beats. This solution is used to estimate new per-beat positions by minimizing forward solution error. The method was tested using data synthesized using measurements from a torso tank phantom with a suspended canine heart. We simulated moving the heart to 100 locations within 40x40x90 mm bounds, computing torso surface measurements per-location and adding noise. Improvement was evaluated in terms of both localization and ECGI accuracy.

Results: The mean per-electrode localization error was 8.4+/-0.2~mm after correction. ECGI accuracy increased using the corrected as compared to uncorrected geometry. Spatial and temporal correlation increased from 53+/-27% to 94+/-0.08%, and 52+/-27% to 94+/-0.7% respectively, and RMSE decreased from 1.1~+/-0.6 mV to 0.33~+/-0.0087 mV.

Discussion: Our geometric correction method dramatically improved ECGI accuracy by reducing cardiac localization error. Future studies will extend to more realistic animal models and then human subjects. Success could impact clinical ECGI by reducing errors from respiratory movement and perhaps dramatically reduce imaging requirements and thus cost and difficulty of use, widening clinical applicability.