EEG minimum-norm estimation compared with MEG dipole fitting in the localization of somatosensory sources at S1

Dipole models, which are frequently used in attempts to solve the electromagnetic inverse problem, require explicit a priori assumptions about the cerebral current sources. This is not the case for solutions based on minimum-norm estimates. In the present study, we evaluated the spatial accuracy of the L2 minimum-norm estimate (MNE) in realistic noise conditions by assessing its ability to localize sources of evoked responses at the primary somatosensory cortex (SI).

Multichannel somatosensory evoked potentials (SEPs) and magnetic fields (SEFs) were recorded in 5 subjects while stimulating the median and ulnar nerves at the left wrist. A Tikhonov-regularized L2-MNE, constructed on a spherical surface from the SEP signals, was compared with an equivalent current dipole (ECD) solution obtained from the SEFs.

Primarily tangential current sources accounted for both SEP and SEF distributions at around 20 ms (N20/N20m) and 70 ms (P70/P70m), which deflections were chosen for comparative analysis. The distances between the locations of the maximum current densities obtained from MNE and the locations of ECDs were on the average 12-13 mm for both deflections and nerves stimulated. In accordance with the somatotopical order of SI, both the MNE and ECD tended to localize median nerve activation more laterally than ulnar nerve activation for the N20/N20m deflection. Simulation experiments further indicated that, with a proper estimate of the source depth and with a good fit of the head model, the MNE can reach a mean accuracy of 5 mm in 0.2-microV root-mean-square noise.

When compared with previously reported localizations based on dipole modelling of SEPs, it appears that equally accurate localization of S1 can be obtained with the MNE.

MNE can be used to verify parametric source modelling results. Having a relatively good localization accuracy and requiring minimal assumptions, the MNE may be useful for the localization of poorly known activity distributions and for tracking activity changes between brain areas as a function of time.