Abstract
In the responses of magnetoencephalography (MEG) related to higher brain functions, components that result in a failure of the estimation of single equivalent-current dipoles often appear. For example, a single dipole source with a very deep location and a very large moment is calculated. This is unacceptable for MEG researchers, though it satisfies the criteria for inspecting the reliability of the dipole, such as goodness-of-fit (GOF) and correlation coefficient. We have developed a new algorithm to avoid these false solutions and to obtain better accuracy of source locations. In the proposed method, a term including the moment and signal-to-noise ratio (SNR) was added to the cost function of the rotating dipole estimation. Two dipoles were successfully localized from the simulated MEG responses that gave an unacceptably deep solution if the conventional cost function was used. The location error was at a distance of less than 5mm between the estimated and original dipoles when the SNR of the response was larger than 10. By use of the present method, the possibility of getting unacceptable solutions from the actual MEG responses will be reduced.
