To estimate potential neural generators, we projected data of each trial into source space using the linearly constrained minimum variance (LCMV) beamformer, which estimates brain activity separately for each source location by minimizing the variance of the filtered signal with the unit-gain constraint 75. The LCMV is widely used for source reconstruction 76 because it provides higher spatial resolution compared to linear inverse methods 77, 78. Source estimation was facilitated by generating a realistically shaped, sufficiently detailed, volume conduction head model comprised of different tissues classes for white (0.14 S/m) and gray (0.33 S/m) matter, compact (0.0064 S/m) and spongy (0.02864 S/m) bone, and cerebral spinal fluid (1.79 S/m) and skin (0.43 S/m) 79, 80 using CURRY 7.0 multimodal neuroimaging software (Compumedics Neuroscan, Charlotte, NC). This 5-layer approach was recently shown to reduce considerable inverse, connectivity, and topographical errors when computing connectivity in source space, as compared with conventional 3-layer models of skin, skull, and brain 81. A finite element method, which allows for modeling arbitrary anisotropic conductivity profiles, was used to solve the EEG forward problem 82–84. Leadfields were
