A comparison of adaptive and non-adaptive EEG source localization algorithms using a realistic head model

An accurate and robust electroencephalogram (EEG) source localization algorithm would be a definite asset for the surgical treatment of patients with epilepsy. Due to the underdetermined nature of the EEG inverse problem, a variety of algorithms with unique constraints and assumptions are applied to select the current dipole source distribution that best accounts for the scalp recordings. We investigated four algorithms: two non-adaptive algorithms: the minimum norm and LORETA as well as two adaptive algorithms: the Borgiotti-Kaplan and eigenspace projection beamformers. Compared over a range of SNR values and single source locations, we found that the eigenspace projection beamformer exhibited superior localizing capabilities compared to the other three algorithms while minimizing source current dispersion. The size of the data window required to accurately localize using the adaptive beamformers was also investigated to improve algorithm efficiency and minimize stationary source assumptions.