Estimating neural sources using a worst-case robust adaptive beamforming approach

Recently, brain source localization and beamforming methods have played an important role in enhancing the utility of the electroencephalograph (EEG) and/or the magnetoencephalograph (MEG). Source localization methods are in general very sensitive to parameter values used to describe the underlying lead–field matrix, such as head shape, electrode positions, conductivity of various tissues of the head, etc. Errors in these parameter values can cause significant degradation in performance of these algorithms. In this paper, we develop a robust minimum variance beamformer (RMVB) specifically for EEG applications that can deal with an arbitrary mismatch between the assumed and true lead field matrix. The approach optimizes the worst-case uncertainty performance, sacrificing a distortionless response in exchange for more robust performance. The performance of the RMVB is compared with the classic minimum variance beamformer (MVB) as well as its regularized and eigenspace-based variations. The simulation scenarios highlight the superior ability of the RMVB to cope with arbitrary mismatches.