Ictal localization by source analysis of infraslow activity in DC-coupled scalp EEG recordings

Abstract

New bedside long-term DC-coupled EEG techniques have demonstrated that infraslow (< 0.5 Hz) activity lateralizes temporal lobe seizures (Vanhatalo, S., Holmes, M.D., Tallgren, P., Voipio, J., Kaila, K., Miller, J.W., 2003a. Very slow EEG responses indicate the laterality of temporal lobe seizures: a DC-EEG study. Neurology 60, 1098–1104). However, even high amplitude infraslow activity is difficult to localize by simple visual inspection if there is overlying faster EEG activity or slow artifact. In this study, we address this with improved DC-coupled EEG recording and analysis techniques and also extend observation to both temporal and extratemporal seizures. Recordings were performed during presurgical evaluation of medically intractable epilepsy, with 20 seizures in 11 patients analyzed. A commercial DC-coupled recording device was used, with sintered Ag/AgCl electrodes in a standard 10-10 system array, with additional anterior temporal and subtemporal electrodes. Seizures were localized with a software package by means of source montage analysis. Infraslow signals occurred with all seizures, often with amplitude orders of magnitude higher than conventional frequencies (0.5 to 70 Hz). The most reliable method to localize these signals and distinguish them from artifacts used a source montage after low-pass filtering below 0.5 Hz. Five of the eight patients who received epilepsy surgery had follow-up documenting significant seizure reduction, and infraslow signal analysis correctly localized the region of seizure onset in all five, while conventional noninvasive EEG recording and analysis localized only three of the five. Several seizures were also analyzed using principle component analysis source localization methods, with the results less consistently localizing than source montage analysis. DC-coupled EEG recordings give clinically useful information to noninvasively localize the seizure focus. The value of this method is increased by source analysis tools that reveal localized changes more clearly than direct visual inspection.

Introduction

Neurosurgical removal of epileptogenic tissue is an important treatment for medically intractable localization related epilepsy. The surgical target is identified by correlating different diagnostic modalities, including electroencephalographic (EEG) recordings, and neuroimaging (Engel, 1993). Recording of spontaneous seizures with scalp (noninvasive) electrodes is a key step but may give equivocal localization (Foldvary et al., 2001, Risinger et al., 1989), so recordings with intracranial electrodes are sometimes needed.

Ictal electrographic discharges in animals and humans (O'Leary and Goldring, 1964, Caspers, 1993, Gumnit, 1974) contain substantial activity at infraslow frequencies (defined as less than 0.5 Hz; Vanhatalo et al., 2004a, Vanhatalo et al., 2004b) in addition to the changes occurring in the conventional EEG bandwidth. Early human studies include intraoperative direct-current (DC)-coupled electrocorticographic recordings with calomel half-cell electrodes (Goldring, 1963) and DC-coupled EEG recording of generalized spike and wave discharges with scalp electrodes (Bates, 1963, Cohn, 1964, Chatrian et al., 1968). The recordings of that era required custom amplifiers needing rebalancing every few minutes to correct baseline drift (Chatrian, personal communication), precluding introduction of this technique into clinical practice. Genuine DC-coupled recordings also require nonpolarizing electrodes (typically Ag/AgCl). Nonetheless, a few papers have studied low-frequency shifts with polarizable electrodes and conventional AC amplifiers. One study (Gross et al., 1999), using stainless steel subdural electrodes and a 0.01-Hz high-pass filter, did not find baseline shifts with most seizures. Another group (Ikeda et al., 1996, Ikeda et al., 1999), using subdural platinum electrodes (which have better low-frequency recording properties; Tallgren et al., 2005) and a high-pass filter of 0.016 Hz, found focal ictal shifts, that were more localized than higher frequency ictal phenomena, and also reported low-frequency baseline shifts with scalp Ag/AgCl electrode recordings in three patients during frontal and parietal seizures. This work (Ikeda et al., 1999) suggested that ictal infraslow signals could be of value for localization in the presurgical evaluation.

Recently, techniques have been perfected for long-term, stable, bedside DC-coupled EEG recordings from human scalp (Bauer et al., 1989, Vanhatalo et al., 2002, Vanhatalo et al., 2003b, Vanhatalo et al., 2003c). In a prior study (Vanhatalo et al., 2003a), we examined seven patients with temporal lobe epilepsy undergoing presurgical evaluation with a custom-made 16-channel DC amplifier and Ag/AgCl electrodes. All seizures were associated with negative 30–150 μV infraslow shifts over the temporal region at about the same time as higher frequency discharges. The side of the infraslow shift always agreed with other diagnostic tests and, at times, was more clearly lateralized than conventional EEG changes. This (Vanhatalo et al., 2003a) demonstrated that noninvasive DC-coupled EEG recordings are inexpensive, practical and useful, and might reduce need for invasive monitoring. This study did not investigate the value of DC-coupled EEG recordings for extratemporal seizures, which are typically even more difficult to localize with conventional scalp EEG (Shukla et al., 2003). The most important limitation of that study (Vanhatalo et al., 2003a), however, was that the infraslow ictal shifts were difficult to localize on direct visual inspection of the recording and occasionally were obscured by overlying faster activity and slow artifacts, such as potentials generated by eye and tongue movements (Vanhatalo et al., 2003a, Vanhatalo et al., 2003c).

The primary goal of the present study was to develop and test new recording and analysis tools to localize ictal infraslow changes. A secondary aim was to extend our observations to a variety of temporal and extratemporal seizure types. We performed long-term DC-coupled EEG recordings with new instrumentation recording up to 27 electrodes. In addition, we incorporated virtual source montage and principal components analysis to identify and localize ictal infraslow signals. Source montage analysis assumes that the electrical activity at the electrode locations on the scalp is a linear transformation of distinct cortical source/current activities (or “source derivatives”) within the brain, and uses an a priori model comprising predefined cortical locations (virtual current sources) within a head topography (Scherg et al., 2002, Ille et al., 2002), to effectively derive an inverse transformation for the scalp electrical activity to source derivative activity. Moreover, this process produces spatial filtering effects that suppress overlapping activities of neighboring sources (Ille et al., 2002) for easier localization of scalp-recorded EEG signals, including the infraslow activity that typically involves several scalp electrodes on conventional montages. This also allows spatial separation of ictal shifts from localized artifacts such as eye movements. Another method that has been used for source localization is principal components analysis (Ille et al., 2002, Lagerlund et al., 1997), which decomposes the scalp EEG channel waveforms into mutually statistically uncorrelated components that most explain the signal variance and then utilizes inverse modeling to determine optimal intracranial locations for these components. In this paper, we applied these methods to determine the most useful approach to analysis of infraslow and conventional bandwidth signals from DC-coupled recording of seizures and to compare ictal EEG activity in three frequency bands. We tested the hypothesis that ictal discharges have highest amplitude at the infraslow frequencies (< 0.5 Hz), and that analysis of these lower frequency signals more clearly localizes the region of seizures onset.