Brain signal source localization using a method combining BP neural networks with nonlinear least squares method

Abstract

Brain source localization is an important inverse problem for brain diagnosis and functional analysis. The goal of dipole source localization in the brain is to estimate a set of parameters that can represent the characteristics of the source. Although a back-propagation neural networks (BPNN) method can solve this typical inverse problem fast enough for real time localization, the accuracy may not be high enough. A problem in using a nonlinear least squares (NLS) method is that the solution may be trapped in the local minima of an error function or be not converged. A method combining BPNN with NLS is proposed in this study. The method shows how to estimate an approximate solution of the inverse problem by the BPNN, and how to select the initial value of the NLS due to the results of BPNN to obtain the optimal solution.