Experimental tests of EEG source localization accuracy in spherical head models

Objectives: The locations of electrical sources in the brain can be calcula ted using EEG data. However, the accuracy of these calculations is not well known because it is usually not possible to compare calculated source loca tions with actual locations since little accurate location information is a vailable about most sources in the brain. Methods: In this study, sources at known locations are created by injecting current into electrodes implanted in the brains of human subjects. The loc ations of the implanted and scalp EEG electrodes are determined from CTs. T he EEG signals produced by these dipolar sources are used to calculate sour ce locations in spherical head models containing brain, skull, and scalp la yers. The brain and scalp layers have the same electrical conductivity whil e 3 different skull conductivity ratios of 1/80th, 1/40th, and 1/20th of br ain and scalp conductivity are used. Localization errors have been determin ed for 177 sources in 13 subjects. Results: An average localization error of 10.6 (SD = 5.5) mm for all 177 so urce was obtained for a skull conductivity ratio of 1/40. The average error s for the other ratios an only a few millimeters larger. The average locali zation error for 108 sources at superior locations in the brain is 9.2 (4.3 ) mm. The average error for 69 inferior location sources is 12.8 (6.2) mm. There are no significant differences in localization accuracy for deep and superficial sources. Conclusions: These results indicate that the best average localization that can be achieved using a spherical head model is approximately 10 mm. More realistic head models will be required for greater localization accuracy. ( C) 2001 Elsevier Science Ireland Ltd. All rights reserved.