Session S32.3

Reconstruction of Transmembrane Currents Using Support Vector Machines and Its Application to Endocardial Mapping: A Model Study

F Alonso-Atienza*, JL Rojo-Álvarez, D Álvarez, M Moscoso, A García-Alberola

Universidad Rey Juan Carlos
Fuenlabrada, Spain

Despite of the inverse problem of electrocardiology (IPE)has been long studied, some of its major problems, such as ill-posing, curse of dimensionality, and clinical validation, are still being paid attention, and many different regularization techniques have been explored in this setting. Support Vector Machines (SVM) are been widely used in many other problems, such as regression or classification, in which they have been shown to have excellent properties of regularization and robustness against the curse of dimensionality, as they base on the Structural Risk Minimization Principle. Therefore, we propose here two new SVM algorithms, specifically adapted to the IPE, and we hypothesize that they can yield better performance than conventional regularization techniques. We focus here on the ill-posing issues of the IPE, and develop the equations for endocardial mapping of transmembrane currents. We show, both in simple simulations and in a previously developed cellular automaton model of the cardiac tissue, that the ill-posing robustness of the SVM is higher when compared to regularized approaches during the depolarization phase. In conclusion, the properties of the developed SVM algorithms stand for an appropriate framework for addressing the IPE.

(Abstract Control Number: 160)