Source localization of EEG versus MEG: empirical comparison using visually evoked responses and theoretical...

Theoretically, the information we can obtain about the functional localization of a source of brain activity from the scalp, for instance evoked by a sensory stimulus, is the same whether one uses EEG or MEG recordings. However, the nature of the sources and, especially of the volume conductor, poses constraints such that appreciable differences between both types of data may exist. We present here empirical and theoretical data that illustrate which are the main constraints and to what extent they may affect electric potential and magnetic field maps. The empirical data consists of visual evoked potential and magnetic fields to the appearance of a checkerboard pattern (half-visual field stimulation). The concept of equivalent dipole is presented and its limitations are discussed. It is considered that the concept of equivalent dipole (ED) yields only an approximate description of the activity of a patch of cortex. A main difference between EEG and MEG recordings is the fact that radially oriented dipoles can hardly be seen in the MEG in contrast with the EEG. Accordingly, a weak tangential dipole component is difficult to distinguish in the EEG if a strong radial component is also present. However, a combination of both methods can give useful complementary information in such cases. A factor that influences largely such differences is the model of volume conductor used. A four concentric spheres model, as commonly used for solving the inverse problem of source localization, causes appreciable errors when EEG data are used but much less in case of the MEG. The use of a model consisting of eccentric spheres fitting the four compartments, brain, CSF, skull and scalp, provides a better approximation of the real geometry of the head and allows to obtain comparable results for visual evoked potentials and magnetic fields. It is emphasized that for precise localization of EDs, especially based on EEG recordings, a realistic model of the different compartments of the head is necessary. The latter must be tailor made to a given subject using MRI-scans, in view of the large variability in head geometry between subjects.