Performance of inverse problem regularization methods for driver location during atrial fibrillation

Locating the atrial fibrillation (AF) sources is a relevant and not fully analyzed problem. We propose a procedure to benchmark methods for driver location in AF and compared three representative techniques: zero-order Tikhonov, Greensite and Bayes (maximum a posteriori). These methods were used to estimate the epicardial potentials, in turn used to locate the driver, using a realistic computer model for atria and torso with two simulated AF propagation patterns. The assessment is based on the spatial mass function of the driver location (SMF), i.e. the probability of the driver being at each point of the atria. Being the driver region (DR) the points with SMF > 0, we defined three metrics: (i) weighted under-estimation indicator, which is the weighted percentage of the true DR that is not detected out of the entire true DR; (ii) the weighted over-estimation indicator, which is the percentage of the misjudged DR out of the entire estimated DR; and (iii) the correlation coefficient between real and estimated SMFs. Results show that the these metrics are easy to compute and provide representative information about the location accuracy. Among the compared algorithms, Bayes method provided the best performance in both AF patterns. Remarkably, even for the most complex pattern, for which epicardial potentials estimation was inaccurate, the three methods approximately located the activity driver.