Abstract
In recent years, magnetoencephalography (MEG) techniques have been developed for estimating internal electrical sources in the human brain from surface measurements of magnetic fields. In this study, we focused on speeding up distributed source estimation, using the sub-optimal least-squares subspace scanning technique with multiple scanning resolutions. As a first step, we set a coarse scanning grid over a large area of the head. On the grid points, we calculated the cost function to be used as the criterion for the existence of a source. Then, as a second step, we set a fine grid on the area with the largest cost function, and calculated the cost function again. We verified the effectiveness of this method by computer simulation, and applied it to MEG data associated with word recognition processes in the human brain. The results showed that the amount of calculation required for the source scanning could be decreased to 1/20 without decreasing the spatial resolution around the source area.
