MEG analysis with spatial filter and multiple linear regression

A spatial filter for MEG analysis which does not utilize any temporal and prior information is proposed. The spatial filter is normalized to satisfy the criterion which is derived from the definition of the spatial filter. Due to the normalization, the spatial filter outputs the largest value at its target position. Furthermore, the current density distribution estimated with spatial filter is localized with Mallows C/sub p/ statistic which selects an optimum regression model. Some numerical experiments verify that this method estimates almost correct positions of dipoles. It is also confirmed that new method we propose gives more reliable estimation than the conventional method which decides dipole on the position of the largest current density estimated with spatial filter iteratively.