Evaluation of smoothing in an iterative lp-norm minimization algorithm for surface-based source localization of MEG

Abstract

The imaging of neural sources of magnetoencephalographic data based on distributed source models requires additional constraints on the source distribution in order to overcome ill-posedness and obtain a plausible solution. The minimum l(p) norm (0 < p < or = 1) constraint is known to be appropriate for reconstructing focal sources distributed in several regions. A well-known recursive method for solving the l(p)-norm minimization problem, for example, is the focal underdetermined system solver (FOCUSS). However, this iterative algorithm tends to give spurious sources when the noise level is high. In this study, we present an algorithm to incorporate a smoothing technique into the FOCUSS algorithm and test different smoothing kernels in a surface-based cortical source space. Simulations with cortical source patches assumed in auditory areas show that the incorporation of the smoothing procedure improves the performance of the FOCUSS algorithm, and that using the geodesic distance for constructing a smoothing kernel is a better choice than using the Euclidean one, particularly when employing a cortical source space. We also apply these methods to a real data set obtained from an auditory experiment and illustrate their applicability to realistic data by presenting the reconstructed source images localized in the superior temporal gyrus.