Source localization

Source localization was performed to obtain the MMN and P300 generators in the brain across various modalities. The EEG data across all sensors were re-referenced with an average reference prior to source localization. Nineteen participants’ structural MRI data were resliced and segmented to identify the brain, skull, and scalp tissues. Two subjects’ MRI scans could not be obtained because of their incompatibility with the fMRI scanner. The origin (0,0,0) of all the T1 images was set to the anterior commissure. Participant-specific headmodel was computed using the OpenMEEG toolbox (Gramfort et al., 2010), using realistic conductivity values. The Polhemus data was imported to place the sensor locations on the head model of each participant. To obtain high accuracy of electrode positions, individual coregistration was employed by firstly visually marking the fiducials (nasion, left preauricular, and right preauricular) in the MRI volume and finally matching the marked points with the fiducial locations as per the Polhemus data. Next, the sources were placed in the segmented part of the head model at a distance of 5 mm from each other and the leadfield matrix was computed i.e., a transfer function between each source point and each electrode. Source localization of each individual was performed using their respective headmodel, leadfield, and electrode positions. For the inverse solution, covariance across all EEG sensors was computed for each condition. eLORETA, belonging to a class of current density measures to calculate the distribution of neuronal sources, was used to solve the inverse problem (Pascual-Marqui, 2007). eLORETA also generates the least amount of false-positives; hence, it is beneficial for exploratory source analysis, for example, where prior hypotheses regarding approximate locations may not be available (Halder et al., 2019). Lambda of 0.1 was used as the regularization parameter for the localization of P300 and MMN ERPs. After localization, each individual's source intensities were interpolated to their respective MRI volume. Further, to calculate the grand average of the source values, the interpolated images were normalized to a common structural template. Finally, we subtracted the voxel intensities of oddball and standard categories and the voxels having intensities >99th percentile were thresholded. This was done separately for each condition and hemisphere.