Global Optimization in the Localization of Brain Activity

The parameters of a multidipole model are usually estimated by finding the values that minimize the sum-of-squares error. Conventional dipole localization usually works well if the number of sources and their approximate locations are known. When data from the whole cortex are analyzed, however, several interacting sources are seen; conventional methods are then often inadequate because the optimization becomes easily trapped in local minima. We studied three different global optimization methods that localize several current dipoles simultaneously without need for any initial guesses.