EEG source localization for two dipoles by neural networks

It is reported that neural networks are useful for solving physiological inverse problems such as EEG and MEG source localizations, where a single dipole source is localized in terms of position and direction. This method is developed here to localize two independent sources from EEG. To improve localization accuracy, 36 electrodes for EEG measurement are used and two dipole sources are assumed to be in each restricted region of the brain. As a result, localization accuracy has been about 3 to 9% depending on the region restrictions, which may be useful for real time epileptic diagnoses.