Epileptic source localization with high density EEG: how many electrodes are needed? G. Lantz a,b, * , R. Grave de Peralta a , L. Spinelli a , M. Seeck c , C.M. Michel a,b a Functional Brain Mapping Laboratory, Department of Neurology, University Hospital of Geneva, 24,Rue Micheli-du-Crest, CH-1211, Geneva, Switzerland b Plurifaculty Program of Cognitive Neuroscience, University of Geneva, Geneva, Switzerland c Laboratory of Presurgical Epilepsy Evaluation, Functional Neurology and Neurosurgery Program of the University Hospitals Lausanne and Geneva, Switzerland Accepted 15 October 2002 Abstract Objective: Electroencephalography (EEG) source reconstruction is becoming recognized as a useful technique to non-invasively localize the epileptic focus. Whereas, large array magnetoencephalography (MEG) systems are available since quite some time, application diffi- culties have previously prevented multichannel EEG recordings. Recently, however, EEG systems which allow for quick (10–20 min) application of, and recording from, up to 125 electrodes have become available. The purpose of the current investigation was to system- atically compare the accuracy of epileptic source localization with high electrode density to that obtained with sparser electrode setups. Methods: Interictal epileptiform activity was recorded with 123 electrodes in 14 epileptic patients undergoing presurgical evaluation. Each single epileptiform potential was down sampled to 63 and 31 electrodes, and a distributed source model (EPIFOCUS) was used to reconstruct the sources with the 3 different electrode configurations. The localization accuracy with the 3 electrode setups was then assessed, by determining the distance from the inverse solution, maximum of each single spike to the epileptogenic lesion. Results: In 9/14 patients, the distance from the EEG source to the lesion was significantly smaller with 63 than with 31 electrodes, and increasing the number of electrodes to 123 increased this number of patients from 9 to 11. Simulations confirmed the relation between the number of electrodes and localization accuracy. Conclusions: The results illustrate the necessity of multichannel EEG recordings for high source location accuracy in epileptic patients. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Multichannel electroencephalograph; Source localization; Epilepsy 1. Introduction It is now recognized that electromagnetic non-invasive source imaging may localize, precisely, primary epileptic foci with high spatial and temporal resolution. To achieve this, and to differentiate primary foci from propagated areas, a sufficient spatial sampling is needed. It has been demon- strated (Bendat and Piersol, 1986) that the sampling rate (both spatial and temporal) has to be at least 2.5 times the highest frequency component of the signal (the Nyquist frequency) in order to avoid aliasing (i.e. contamination from higher frequencies through increase of energy in lower frequency bands). As opposed to temporal sampling, spatial sampling is by definition non-continuous, which makes low pass filtering, in order to avoid aliasing, impos- sible (Srinivasan et al., 1998). Consequently, high spatial sampling is necessary to correctly characterize the topogra- phical details of all frequencies as well as to avoid distortion of frequencies of interest from frequencies above the Nyquist frequency. To some extent, the problem can be diminished by the fact that the scalp acts as a spatial low pass filter, attenuating the undesired high frequency infor- mation (Nunez, 1981, Srinivasan et al., 1996). Even so, recording, for instance with the commonly used 21 elec- trode setup (corresponding to 6 cm inter-electrode distance), is definitely insufficient to adequately sample the spatial frequencies of around 3 cm appearing, for example, in early evoked potential components (Spitzer et al., 1989; Gevins, 1990). In a simulation study by Vanrumste et al. (2000), addressing the effect of volume conductor model errors on EEG dipole source localization, smaller localiza- tion errors were found for 53 rather than for 27 electrodes, even if this difference was quite small. Theoretical and experimental studies by Srinivasan et al. (1998) have Clinical Neurophysiology 114 (2003) 63–69 1388-2457/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S1388-2457(02)00337-1 www.elsevier.com/locate/clinph CLINPH 2002577 * Corresponding author. Tel.: 1 41-22-3728339; fax: 141-22-3728333. E-mail address: goran.lantz@medecine.unige.ch (G. Lantz).