We used a real-skull phantom head to investigate the performances
of representative methods for EEG source localization when considering various
head models.
We describe several experiments using a montage with current sources located
at multiple positions and orientations inside a human skull filled with a
conductive medium. The robustness of selected methods based on distributed
source models is evaluated as various solutions to the forward problem (from
the sphere to the finite element method) are considered.
Experimental results indicate that inverse methods using appropriate
cortex-based source models are almost always able to locate the active source
with excellent precision, with little or no spurious activity in close or
distant regions, even when two sources are simultaneously active.
Superior regularization schemes for solving the inverse problem can dramatically help
the estimation of sparse and focal active zones, despite
significant approximation of the head geometry and the conductivity properties
of the head tissues. Realistic head models are necessary, though, to fit the
data with a reasonable level of residual variance.