Comparison of Algorithms for the Localization of Focal Sources: Evaluation with simulated data and analysis of experimental data.

Abstract

This paper presents a comparative study of the capabilities of five distributed linear solutions to accurately determine the position of single sources. Two recently developed inverse solutions, LAURA and EPIFOCUS are compared to the Minimum Norm, the column Weighted Minimum Norm and the Minimum Laplacian. The comparison is based on three figures of merit: 1) the number of sources with zero localization error, 2) the maximum localization error, and 3) the average localization error as a function of the source eccentricity. The best results in terms of the three figures of merit are obtained for EPIFOCUS and LAURA. We report for the first time a linear inverse solution (EPIFOCUS) capable of localizing all single sources with zero dipole localization error for a relatively low number of sensors (100). The robustness of EPIFOCUS is additionally evaluated in this paper with noisy synthetic data and experimental recordings in epileptic patients. It is concluded that EPIFOCUS is a robust method to localize single sources with the following advantages over single dipole localization: 1) It is computationally more efficient. 2) It is easily applicable to realistic head models (gray matter selected from MRI). 3) Sources are not restricted to be dipolar. The study described in the paper endorses an important theoretical conclusion: While it is possible to design linear solutions with optimal performance in the determination of the position of single sources, such performance is not warranted if multiple sources are simultaneously active. Consequently, lower dipole localization error is neither a sufficient nor a necessary condition for the performance of a linear inverse solution.