Abstract
Magnetoencephalography (MEG) source reconstruction problems suffer from their own ill-posed characteristics, especially when a cortically distributed source model is used. This is because many variables have to be determined using small amounts of measured data, and this leads to a large computational cost. In this paper, a very efficient method is proposed to localize the source space effectively. A Fourier based 3-dimensional spatial function approximates the spatial distribution of the current dipole moments. The coefficients of the function are determined using the sensitivity analysis employing the steepest decent updating scheme. By repeatedly applying the method to smaller and smaller domains, a reduced source space, which includes the small number of variables, can be obtained. By applying this method to a real brain model, its usefulness was verified.