Partitioning of deep versus superficial intracranial sources using current source densities is not valid

An analytic method has recently been proposed for partitioning scalp-recorded EEG and evoked potential (EP) data into parts arising from deep (i.e., subcortical) vs. superficial (i.e., cortical) sources. The method is based on the observation that the current source density (CSD) is selectively sensitive to electrical activity arising from superficial sources, and the conjecture that the residual potential which remains after subtracting the CSD from the scalp potentials, represents activity from deep sources. We investigated the validity of this procedure by simulating scalp potential data for superficial and deep dipole sources with known locations and orientations. Our single-dipole simulations demonstrated that, when the actual location of the source was superficial, the partitioning procedure erroneously attributed a sizeable proportion of the total topographic variance to the activity of deeper sources. This produced a consistent bias in the simulations with two dipoles, when both superficial and deep sources were present. In such cases, the relative contribution of the deeper source was consistently overestimated, and the scalp topography of the deep source activity was profoundly misrepresented by the residual which results from subtracting the CSD from the scalp potential. We conclude that the proposed method for partitioning EEG and EP data into components arising from deep vs. superficial intracranial sources is not valid.