Effect of Geometric Complexity of the Heart Model on the Inverse Problem of Electrocardiography.

The electrocardiographic (ECG) inverse problem offers more sophisticated approaches for assessing the electrical condition of the heart than the standard 12-lead ECG. Because of the ill-posted nature of the ECG inverse problem, regularization must be used to obtain clinically meaningful estimates of heart-surface potentials. We compared optimal inverse solutions using L1 regularization with those found using zero-order Tikhonov (ZOT) regularization in two heart-torso models with known heart-surface potentials, but different spatial detail. In the model based on the Visible Human Male with an epicardial-shell heart-model, there was no difference in relative error (RE) and correlation coefficient (CC) during activation and recovery of the ventricles (QT interval) between L1 and ZOT inferred heart-surface potentials. In the heart-torso model based on the ECGSIM package with a ventricular heart model there were significant differences in RE and CC throughout the QT interval. At the peak of the R wave, RE was reduced from 0.841 (ZOT) to 0.358 (L1). L1 regularization reduced RE compared to ZOT except during the downstroke of the R wave. These results suggest that improvement with L1 regularization is dependent on heart model complexity, not its sampling density. [ABSTRACT FROM AUTHOR]