|
Assuming over-discharge of a focus in the brain as an
equivalent current dipole source, localization of epileptic
focus can be obtained from EEG dipole modeling. Usually, the
location and orientation of the dipoles can be determined by
iterative calculations using optimization algorithm
minimizing a cost function. However, the critical dependence
on the initial estimation is an inherent feature of iterative
dipole optimization algorithms. Improved initial estimation
method for noninvasive localization of epileptic
focus, modeled as electrical dipole, is proposed in this
research. This study was
accomplished by using singular value decomposition (SVD)
technique which has been commonly used for extracting the single
dipole component from multiple dipoles and background noise. In
addition, a 3-diomensional (3-D) representation of EEG maps and
reliable 3-D spline interpolation method makes it possible to
obtain better initial estimation than those with
standard planar mapping. In general, the dipolecomponents are
located at the zero potential plane, called null plane.
However, for 3D topographic mapping, many zero potential points
can be detected. Instead of testing all the possible points,
the proposed approach is to fit the selected zero
potentials in terms of a linear plane, i.e. null plane, via
least square approach. The initial estimates of the parameters
which determine the dipole modeling can be computed from the
intersection between null plane and the line connecting the
peak and valley potentials.
For verifying the proposed approach, the EEG potential simulated
from current dipoles in an ideal homogeneous medium are
generated. The localization discrepancy and the moment
orientation discrepancy are used as index for measuring the
estimation accuracy. The distribution of initial estimation
discrepancy was further modeled as a mathematical gamma
function so that distribution features could be characterized.
The initial estimation of various simulation data, including
noise data, noise data with SVD, and noise free data are
compared. For clinical EEG, subjects with different layer such
as mesiotemporal lobe and temporal lobe epilepsy are
compared. Our experimental results indicate that having
good initial estimates for the dipole parameters is essential to
ensure rapid convergence to the correct solution.
|