Neural dipole localization by a hybrid nonlinear optimization algorithm

Abstract

In the MEG inverse problem, the source localization procedure is to obtain dipole parameter solution that produces a calculated field pattern best matching the measured data. Here, a hybrid algorithm is described, i.e., Levenberg-Marquardt (LM) method for a fine scanning near the source area, and quasi-Newton (QN) method for a high-speed coarse scanning over a large area of the head. By a set of simulations, this presented algorithm can be more efficient both in computation time and sensitivity to the iterative initial value.