Fast EEG/MEG BEM-based forward problem solution for high-resolution head models

Abstract

• A boundary element method with fast multipole method acceleration (BEM-FMM) is applied to solve an EEG forward problem on detailed high-resolution human head models within approximately 60 seconds on a standard workstation. • The BEM-FMM is augmented by a fast one-time adaptive mesh refinement method ( b -refinement) that brings solution error below 5% when compared against an analytical solution (concentric spherical shells test case) and a solution carried out with a full adaptive h -refinement method (1 M facet head models test case). • The b -refinement method shows a 30x speedup relative to the h- refinement method while producing equivalent results and increasing the model size by only 10-15% on average. • The method is applicable to EEG source localization problems based on measured experimental EEG data. A fast BEM (boundary element method) based approach is developed to solve an EEG/MEG forward problem for a modern high-resolution head model. The method utilizes a charge-based BEM accelerated by the fast multipole method (BEM-FMM) with an adaptive mesh pre-refinement method (called b -refinement) close to the singular dipole source(s). No costly matrix-filling or direct solution steps typical for the standard BEM are required; the method generates on-skin voltages as well as MEG magnetic fields for high-resolution head models within 90 seconds after initial model assembly using a regular workstation. The forward method is validated by comparison against an analytical solution on a spherical shell model as well as comparison against a full h -refinement method on realistic 1M facet human head models, both of which yield agreement to within 5% for the EEG skin potential and MEG magnetic fields. The method is further applied to an EEG source localization (inverse) problem for real human data, and a reasonable source dipole distribution is found.

Topics & Concepts