Combination of Computer Simulations and Experimental Measurements as the Training Dataset for Statistical Estimation of Epicardial Activation Maps From Venous Catheter Recordings

Cunedioglu U., Yilmaz B.

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, vol.56, no.3, pp.837-845, 2009 (SCI-Expanded) identifier identifier identifier


One of the epicardial mapping techniques requires the insertion of multiple multi-electrode catheters into the coronary vessels. The recordings from the intracoronary catheters reflect the electrical activity on the nearby epicardial sites; however, most of epicardial surface is still inaccessible. In order to overcome this limited access problem, a method called the linear least squares estimation was proposed for the reconstruction of high-resolution maps using sparse measurements. In this technique, the relationship between catheter measurements and the remaining sites on the epicardium is created from previously obtained high-resolution maps (training dataset). Even though open-chest surgery is still a relatively frequent occurrence, an additional burden on the patient to obtain epicardial maps might impose an important risk on the patient. In this study, we hypothesize that epicardial maps created from computer simulations might be used in combination with the experimental data. In order to test this hypothesis, we used high-resolution epicardial activation maps acquired from 13 experiments performed on canine hearts that were stimulated via unipolar pacing from sites distributed all over the epicardium. We investigated the feasibility of the Aliev-Panfilov model that generated focal epicardial arrhythmias on Auckland heart. We started the simulations from the sites that corresponded to the pacing sites on the experimental geometry after a registration procedure between the experimental and simulation geometries. We then compared the simulation results with the corresponding experimental activation maps. Finally, we included simulated activation maps alone (100%) and in combination (simulated maps constituted 90%, 75%, 50%, 25%, 10%, and 0% of the training dataset) with experimental maps in the training set, performed the statistical estimation, and obtained the error statistics. The mean correlation coefficient (CC) between the simulated epicardial activation maps with the experimental maps was 0.88. The results of the estimation indicated that only 2.6 mm localization error and 0.05 CC value degradation occurred on average for the replacement of 75% of the experimental maps with the simulated counterparts. This indicated that including an important percentage from simulations may lead to decreased need for open-chest procedures and less burden on the patients.