Dipole source localization by means of maximum likelihood estimation. I. Theory and simulations

By analyzing simulated neuromagnetic recordings with a 37 channel magnetometer system, it was shown that the accuracy of a dipole source localization is considerably improved if the standard least-squares fit procedure is replaced by a maximum likelihood estimation accounting for the covariances of the noise in the measurement channels. Spatially correlated noise was generated by random dipoles homogeneously distributed in a sphere representing the brain. The study suggests that a maximum likelihood estimation reduces the standard deviations of the estimated dipole parameters by roughly a factor of two.