Spatiotemporal EEG Analysis and Distributed Source Estimation in Presurgical Epilepsy Evaluation

Abstract

In the attempts to localize electric sources in the brain on the basis of multichannel EEG and/or MEG measurements, distributed source estimation procedures have become of increasing interest. Several commercial software packages offer such localization programs and results using these methods are seen more and more frequently in the literature. It is crucial that the users understand the similarities and differences of these methods and that they become aware of the advantages and limitations that are inherent to each approach. This review provides this information from a theoretical as well as from a practical point of view. The theoretical part gives the algorithmic basis of the electromagnetic inverse problem and shows how the different a priori assumptions are formally integrated in these equations. The authors restrict this formalism to the linear inverse solutions i.e., those solutions in which the inversion procedure can be represented as a matrix applied to the data. It will be shown that their properties can be best characterized by their resolution kernels and that methods with optimal resolution matrices can be designed. The authors also discuss the important problem of regularization strategies that are used to minimize the influence of noise. Finally, a new kind of inverse solution, termed ELECTRA (for ELECTRical Analysis), is presented that is based on constraining the source model on the basis of the currents that can actually be measured by the scalp recorded EEG. The practical part of the review illustrates the localization procedures with different clinical data sets. Three aspects become important when working with real data: 1) Clinical data is usually far from ideal (limited number of electrodes, noise, etc.). The behavior of inverse procedures in such unfortunate situations has to be evaluated. 2) The selection of the time points or time periods of interest is crucial, especially in the analysis of spontaneous EEG. 3) Additional information coming from other modalities is usually available and can be incorporated. The authors are illustrating these important points in the case of interictal and ictal epileptiform activity. Spike averaging, frequency domain source localization, and temporal segmentation based on electric field topographies will be discussed. Finally, the technique of EEG-triggered functional magnetic resonance imaging (fMRI) will be illustrated, where EEG is recorded in the magnet and is used to synchronize fMRI acquisition with interictal events. The analysis of both functional data, i.e. the EEG in terms of three-dimensional source localization and the EEG-triggered fMRI, combines the advantages of the two techniques: the temporal resolution of the EEG and the spatial resolution of the fMRI.

Temporal and spatial determination of EEG-seizure onset in the frequency domain

“…The FFT-Approximation method provides ± in contrary to the analysis of power maps ± voltage maps with positive and negative values (as the raw EEG-data). This method as been applied in different data sets (Michel et al, 1992(Michel et al, , 1995Dierks et al, 1993;Lehmann et al, 1993) including EEG of epileptic patients Michel et al, 1999). Fore each seizure in each patient, the 7 FFT-approximated epochs following the ESO were averaged.…”

“…This procedure was carried out, because special coverage of certain brain areas (as used over the temporal lobes during the presurgical evaluation in our patients) leads to a nonuniform coverage of the whole scalp and can disrupt proper source localization . Finally, the surface maps corresponding to the dominant and consistent frequencies were used to localize the source of each frequency map using a linear inverse solution algorithm: Radially Weighted Minimum Norm (Grave de Peralta and Gonzalez, 1998;Michel et al, 1999). The use of the linear inverse solution will allow to control that the analyzed frequency is generated approximately within the brain region whose resection rendered all patients seizure-free.…”

“…We choose the Radially Weighted Minimum Norm solution because it avoids the undesirable tendency of the Minimum Norm solution to reach maxima only at the borders of the solution space while re¯ecting (from a physical point of view) the boundary character of the current density vector. Besides its simplicity this solution has shown a performance comparable or better than more complex solutions in the analysis of epileptic data and evoked potentials (Grave de Peralta, 1999;Michel et al, 1999). The solutions were calculated on a regular grid of 1152 nodes, using a threeshell spherical model.…”

“…3a±i). However more exact methods which use individual realistic head shapes, realistic head models and more complex linear inverse solutions will be necessary to distinguish between the ictal implication of lateral and mesial temporal structures (Fuchs et al, 1999;Michel et al, 1999). In patient 10 the EEG was characterized by strong eye movement artifacts which strongly in¯u-enced the localization procedure for both dominant frequencies (3.0 and 6.0 Hz, see Fig.…”

“…These ®ndings are clinically important since they might help to non-invasively identify patients suffering from seizures arising in the temporal neocortex, because they have less favorable outcomes with standard partial anterior temporal lobectomy. As discussed elsewhere the combination of multiple linear and non-linear inverse solutions as well as realistic head models seems mandatory in order to use a localization method to its full potential (Ebersole, 1999;Fuchs et al, 1999;Michel et al, 1999).…”