EEG Source Reconstruction Using Global Optimization Approaches: Genetic Algorithms versus Simulated Annealing

Abstract

The problem of EEG source localization can be considered as an optimization problem. Various optimization methods were applied to the brain source localization. In this paper we compare the performance of two global optimization techniques, namely genetic algorithms (GA) and simulated annealing (SA), for the estimation of dipole parameters. We found the genetic algorithm approach to be 4 -6 % more effective than simulated annealing regarding convergence to the true global minimum for source reconstruction problems simulated in this work. However, the computational cost of the GA was higher than of the SA. The effectiveness of the mentioned methods and their computational costs were tested and demonstrated through computer simulations with artificial data. We introduce these approaches for brain source localization in the case of a 3-D real head model. The realistic shapes of head tissues were derived from a set of 2-D magnetic resonance images (MRI) by extracting surface boundaries for the major tissues such as the scalp, the skull, the cerebrospinal fluid (CSF), the white matter, and the gray matter. A Finite-element method (FEM) has been used for the modeling of the potentials for an arbitrary complexity head shape. A new objective function obtained from the FEM discretization scheme of the Poisson’s equation with the Neumann boundary condition has been proposed.