The 6th International Congress of the International Society for Brain Electromagnetic Topography (ISBET), Tokushima, Japan

Abstract

EEG and MEG measurements are non-invasive methods containing valuable information about intracranial electrical processes. However, due to the non-uniqueness of the bioelectromagnetic inverse problem, it is by no means straightforward to reconstruct the sources from the measured data. The non-uniqueness has to be overcome by introducing into the model additional knowledge or sensible assumptions about the expected sources. The correctness of the source model is of crucial importance for the reconstruction. Three different bioelectromagnetic reconstruction methods are presented. (1) Non-linear multiple dipole fit assumes that the sources can be described by a small and a priori known number of current dipoles. (2) The MUSIC algorithm does not need the actual number of dipoles, but only the number of independent components in the signal. It also avoids the non-linear optimization procedure. (3) Distributed source reconstruction is based on linear estimation theory and does in principal allow any current source distribution. However, the minimum-norm condition has to be used to keep the solution unique. Alternatives to this rather artificial condition are shown. The advantages and shortcomings of the methods are discussed in detail. Somatosensory evoked potentials (stimulation of the median nerve) were used as a test case. Sources near the hand area of the contralateral somatosensory cortex, as known from literature, are found by all methods. Activity on the ipsilateral side, in the second somatosensory areas, and in the frontal lobes is also found, with great diversity between the methods. We conclude that the inambigious reconstruction of brain electrical sources requires source models the properties of which are completely based on knowledge rather than assumptions.