MULTIPLE-DIPOLE SOURCE LOCALIZATION BASED ON INFORMATION CRITERIA

An information criterion, the Structural Risk Minimization (SRM) princ iple, was applied to magnetic field data obtained from a model of the human brain in which equivalent current dipoles represent active neura l tissues. This criterion was used to estimate the number of active ar eas. Simulation studies were performed on a spherical model of the hea d to evaluate the effectiveness of the SRM principle as such a criteri on, in the following way: 1) the number of the dipoles underlying simu lated magnetic field data was estimated, and 2) attempts were made to distinguish dipoles placed close together. Results were compared with those produced using criteria, such as Akaike Information Criterion (A IC) and the Minimum Description Length (MDL) principle. With SRM, the number of dipoles was estimated correctly for the data generated by as many as 4 dipoles, and it was possible to distinguish dipoles as clos e together as 18 mm. These results were much better than those obtaine d with the MDL or with AIC.