Source localization accuracy in MEG using the Galerkin method

We studied the source localization accuracy in MEG with single-surface brain-shaped conductor models. The Galerkin method with linear basis functions was used as a discretization method. To Þnd a dipolar source, a variant of the Newton method was used for solving nonlinear least-squares problems. Reference magnetic Þelds were computed using a very large number of unknowns to obtain accurate solutions for source localization comparisons. We studied the effect of the number of unknowns on the source localization accuracy, especially near the surface of the brain. We compared the source localization results obtained by the Galerkin method with the results obtained by the collocation method with the same basis functions. Finally, we studied the spatial variation of the relative error of the magnetic Þeld in the temporal lobe.