Address correspondence to Mark D. Holmes, Regional Epilepsy Center, Department of Neurology, University of Washington, Harborview Medical Center, Box 359745, 325 Ninth Avenue, Seattle, Washington 98104, U.S.A. E-mail: email@example.com
Dense array EEG is a method of recording electroencephalography (EEG) with many more electrodes (up to 256) than is utilized with standard techniques that typically employ 19–21 scalp electrodes. The rationale for this approach is to enhance the spatial resolution of scalp EEG. In our research, dense array EEG is used in conjunction with a realistic model of head tissue conductivity and methods of electrographic source analysis to determine cerebral cortical localization of epileptiform discharges. In studies of patients with absence seizures, only localized cortical regions are involved during the attack. Typically, absences are accompanied by “wave–spike” complexes that show, both at the beginning and throughout the ictus, repetitive cycles of stereotyped, localized involvement of mainly mesial and orbital frontal cortex. Dense array EEG can also be used for long-term EEG video monitoring (LTM). We have used dense array EEG LTM to capture seizures in over 40 patients with medically refractory localization-related epilepsy, including both temporal and extra temporal cases, where standard LTM failed to reveal reliable ictal localization. One research goal is to test the validity of dense array LTM findings by comparison with invasive LTM and surgical outcome. Collection of a prospective series of surgical candidates who undergo both procedures is currently underway. Analysis of subjects with either generalized or localization-related seizures suggest that all seizures, including those traditionally classified as “generalized,” propagate through discrete cortical networks. Furthermore, based on initial review of propagation patterns, we hypothesize that all epileptic seizures may be fundamentally corticothalamic or corticolimbic in nature. Dense array EEG may prove useful in noninvasive ictal localization, when standard methods fail. Future research will determine if the method will reduce the need for invasive EEG recordings, or assist in the appropriate placement of novel treatment devices.
The paroxysmal EEG signals recorded on the scalp that reflect the abnormal behavior of neuronal populations in epilepsy have remained for over a half-century as the critically important laboratory findings in the clinical evaluation of patients with epileptic seizures (Niedermeyer, 1999). Despite the indispensable role of the EEG, the standard assessment of electrographic patterns has significant limitations. The usual method of recording from the scalp involves the placement of only 16–21 electrodes that are mainly positioned over the upper portions of the cranium. These circumstances result in distances of several cm separating the electrodes, and examination of basal brain regions that is woefully inadequate. Detailed studies on the spatial frequency spectrum suggests that to maximize spatial information of the human EEG (“spatial Nyquist”), interelectrode distances on the cortical surface must be within 1.25 mm (Freeman et al., 2000), and on the scalp, less than 10 mm (Freeman et al., 2003). As a consequence, standard recordings yield not only poor spatial resolution, but may fail to detect significant pathology. Furthermore, visual analysis of epileptiform discharges from standard recordings provides limited insight into the extent of the involved cortical network, or into the precise patterns of discharge propagation. It is probable that detailed knowledge of these factors, when they become available, will prove to be important both in the understanding of the nature of an individual subject's seizures and in defining appropriate therapy (Spencer, 2002).
Application of major technological advances to the analysis of scalp EEG is now possible and will likely change the current state of affairs. One of these advances is the capability for rapid application of a dense array of electrodes, a technique that may also now be employed for continuous long-term EEG video monitoring (LTM). As many as 256 channels may be employed, with recording systems that include coverage of face and neck in order to noninvasively sample as much electrographic data from basal brain regions as is feasible. Dense array EEG, by increasing spatial sampling and decreasing the distance between electrodes, results in markedly improved spatial resolution from scalp EEG data (Lantz et al., 2003). Another important technological advance is in the development of physical models of head tissues that allow estimation of neural sources of the EEG (Michel et al., 2001). When superior spatial resolution is combined with sophisticated methods of source analysis, with results registered with anatomical features from magnetic resonance imaging (MRI), more precise information on epileptiform electrographic pathology may be gleaned from the scalp EEG (Lantz et al., 2001; Phillips et al., 2002; Michel et al., 2004).
With this background in mind, the purpose of this review is to describe some of the research in dense array EEG, as it applies to epilepsy. The review will conclude with discussions of the insight that this technique may provide in understanding the nature of epileptic circuits, the potential role in the clinical evaluation of difficult seizures, and likely future technological developments.
Dense array EEG recordings
A 256-channel Geodesic Sensor Net is applied to each person during the recordings, requiring 10–30 min for application and adjustment. The net is constructed to include electrode coverage over the face and neck, the purpose of which is to feasibly electrographically sample basal brain regions (e.g., inferior frontal and basal temporal areas). For an average adult head, interelectrode distances are approximately 20–25 mm. The EEG-amplifier characteristic included a band-pass of 0.1–400 Hz and sampling rate of 1000 Hz. In our research on subjects with refractory generalized seizures, recordings are performed on an outpatient basis, with recording times were approximately one hour and with no reduction in subjects' antiseizure medications. LTM with dense array EEG is now possible as well, and we have employed this technique in recording seizures in patients with localization-related seizures, with continuous recordings up to 96 h.
The 256 channel EEG is recorded with a common vertex reference, and rereferenced digitally to various montages for inspection, including the average reference. Because of the improved coverage of the inferior head surface, the average reference allows the potential at each index electrode to be examined with reference to an estimate of the zero potential of the head (Bertrand et al., 1985; Dien, 1998; Junghofer et al., 1999). The average-referenced EEG waveforms are examined with topographic waveform plots, a technique that allows inspection of geometric distribution of the potential fields. In addition, topographic maps are created with spherical spline interpolation (Perrin et al., 1987). Dynamic scalp topography of epileptiform discharges with animations can be created at variable time intervals (Tucker et al., 1994).
Solving the “inverse problem”
The fundamental goal of our research efforts in dense array EEG is to noninvasively localize brain regions that are involved in the onset and distribution of epileptiform discharges. Stated another way, our efforts are aimed at solving the “inverse problem,” that is, determining the location of the origin of electrical signals, in this case from the brain, that are recorded on the surface a closed structure (i.e., the head). However, since there is no unique solution to the inverse problem, the research begins with the construction of a model (or “forward model”) that is based on biologically realistic assumptions, such as the geometry and the electrical properties of the tissues of the head, in order to provide viable answers (Nunez & Srinivasan, 2006). Another assumption, not unexpectedly, is that EEG signals originate form the cerebral cortex. In brief, the method of solving the inverse problem involves use of an appropriate forward model in conjunction with an inverse algorithm for source analysis that is applied to the scalp-recorded EEG data. It is important to emphasize that, regardless of the sophistication of the source analysis, the final results may be misleading if the scalp data lacks adequate spatial resolution. For example, we have shown that in subjects with known temporal lobe epilepsy, the sources of ictal EEG patterns may appear to be entirely outside of the temporal lobe if only standard 21 channel EEG recordings are used.
EEG source analysis
As first step in this analysis, one technique of constructing the forward model is specification of an ellipsoidal head with four homogeneous shells: brain, cerebrospinal fluid (CSF), skull, and scalp. Conductivity ratios that may be used include 1.0 (CSF), 0.3300 (brain), 0.0042 (skull), and 0.3300 (scalp); tissue thicknesses may be specified as 1.0 mm (CSF), 7.0 mm (skull), and 6.0 mm (scalp), with head radius set to 92.5 mm (Berg & Scherg, 1994). An improvement in the forward model may be accomplished by inclusion of a realistic model of head conductivity with finite difference computations (Salman et al., 2005). To provide solutions that are consistent with source analysis techniques regional sources are selected (with dipole moments in three orthogonal directions of space) that are most adequate to describe the discharge complex. Dipole locations are visualized in relation to a standard brain MRI model, with electrodes positioned in relation to skull landmarks in accordance with the international (10–20) EEG system (nasion, periaurcular points), and coregistered with the head conductivity model. The positions of the electrodes, with respect to the standard MRI model, are determined by fitting actual 256-channel locations used in the source localization software. These locations are the average cartesian coordinates of the digitized locations from five normal adult subjects.
Three independent methods have been employed in our research in regard to the inverse algorithm component to identify electrographic sources: two linear inverse techniques and one equivalent dipole method. The linear inverse method of local autoregressive average (LAURA) weights the solution distribution so that sources are continuous with nearby activity (Grave de Peralta et al., 2004). Because the vector fields of electrical sources fall off with the cube of distance (and the potential fields with the square of distance), the LAURA method constrains the solution with a weighting function that assumes the result will have a spatial smoothness with this physical property. The LAURA inverse solutions are implemented within the geodesic electrical source imaging (GESI) software package used in the majority of our research (http://www.egi.com), using the head conductivity model and the probabilistic cortical gray matter locations from the Montreal Neurological Institute (MNI) probabilistic atlas (http://www.bic.mni.mcgill.ca) to constrain the location of 2,400 source voxels on the standard MRI. Three orthogonal dipole moments (x, y, z) are defined and solved for each of the source voxels.
The second linear inverse technique to provide source estimates of the three dimensional cortical distributions of current source density is low-resolution electromagnetic tomography (LORETA) (Pasqual-Marqui et al., 2002). This algorithm solves the inverse problem assuming related orientations and strengths of neighboring neuronal sources, without assuming a specified number of sources. Mathematically, this is solved by finding the “smoothest” of all possible activity distributions. LORETA is registered to the Talairach Brain Atlas (Talairach & Tournoux, 1988) and computes at each voxel of the probabilistic cortical location the current density as the linear weighted sum of the scalp electrical potentials.
The equivalent dipole method to model epilpetiform components is accomplished by means of the brain electrical source activity software (BESA) (Scherg & Ebersole, 1994; Scherg et al., 2002). Dipole location, orientation, amplitude are estimated, with “goodness of fit” derived by residual variance, defined as the percentage of the spike variance that cannot be explained by the model. For standardization, the residual variance is calculated at the midpoint of the rising slope of the epileptiform discharge. Both BESA and LORETA algorithms have been used in our research in conjunction with the ellipsoidal head models.
Dense Array EEG and “Generalized” Seizures
Studies in absence
The conventional classification of epileptic seizures is based on the International League Against Epilepsy dichotomy that epileptic seizures are either “partial” (localization-related) or “generalized” (Commission on Classification and Terminology, 1989). This scheme implies that partial seizures have discrete focal origins, while generalized seizures are assumed to occur without lateralizing or localizing features that include bilateral, global cortical activation at ictal onset (Panayiotopoulos, 2002). Experimental evidence that implicates thalamic and thalmocortical mechanisms in the pathophysiology of generalized seizures has been offered as an explanation for the apparently “generalized” nature of these seizures (Gloor, 1978; McCormick, 2002; Slaght et al., 2002).
The absence seizure is often considered the prototypic idiopathic generalized seizure. Although the concept of generalized seizures persists in clinical practice, traditional EEG visual analysis emphasized a frontal preponderance of the spike–wave complexes in absence (Niedermeyer, 1999) and early studies in source analysis suggested that, although bilateral at onset, the ictal patterns in absence localized to frontal cortex (Rodin et al., 1994). Research using dense array EEG provides further evidence that the traditional concept of “generalized” epilepsy may not be accurate. Recent studies of absence in five patients using 256 channel EEG recordings showed that both at onset and during propagation, the discharges in absence are associated with activation of only discrete regions of mainly medial frontal and orbital frontal cortex (Holmes et al., 2004). Detailed studies of the ictal discharges in absence, the onsets of which develop rapidly, suggest that the waveforms may best be described as “wave–spike,” rather than “spike–wave” (Fig. 1). Though individual variability between subjects may exist, typically the initial slow wave follows oscillations localized to medial frontal regions, then exhibits anterior propagation and abrupt transition over the frontal pole as a positive spike displaces the diffuse anterior negative slow wave, with the discharge then sweeping posteriorly along the orbital frontal cortex. Similar patterns of activation were found using both BESA and LORETA methods of source analysis (Holmes et al., 2004).
Separately, Tucker et al. (2007) analyzed the same data collected for the Holmes et al. (2004) study using the LAURA inverse, the improved high-resolution finite difference head conductivity model, and a “flatmap” display of the cortical surface with GeoSource software (Electrical Geodesics, Inc., Eugene, OR, U.S.A.). Overall, the results were convergent with the earlier report. For each seizure in each patient the slow wave of the wave–spike cycle engaged networks of mainly medial frontal, and occasionally temporal, cortical networks. Invariably, this was followed by primary current source distribution in ventromedial frontal cortex during the abrupt wave–spike transition (Fig. 2). Although differences were found between individual patients, particularly during in the slow wave and the oscillatory EEG changes in the second or so prior to ictal onset, each seizure rapidly progressed to a stereotyped pattern with major spike discharges localized to midline frontal networks. In another report, a series of twelve patients with the idiopathic generalized epilepsy syndrome of juvenile myoclonic epilepsy, Rozas Latorre et al. (2005) also found that the epileptiform discharges in all cases included activation of ventromedial frontal cortex.
Do findings in absence lead to greater understanding of mechanisms of consciousness?
The absence seizure reflects a specific disruption of attention, with momentary impairment of voluntary control of consciousness and capacity to organize action and respond to environmental stimuli. However, continuity of experience and orientation to context is remarkably spared, despite the disruption of conscious attention (Lopes da Silva et al., 2003). Although data from dense array EEG studies reflects only cortical areas activated during absence, we may infer that the ventromedial frontal localization of the spike discharges yields important clues both to the pathophysiology of absence and more generally to the corticothalamic control of over properties of the voluntary regulation of conscious state (Tucker et al., 2007). Animal studies of spike–wave seizures have emphasized pathology of corticothalamic circuits that include the thalamic reticular nucleus (TRN), a thin sheet of cells surrounding the dorsal thalamus (Futatsugi & Riviello, 1998; Steriade, 2003). The TRN, with its anatomical connections, is positioned to serve as a gatekeeper for thalamocortical regulation of cortical activity (Fig. 3). Zikopoulos & Barbas (2006) have shown in primates unique projections from orbital frontal cortex to TRN, interdigitating with sensory input to TRN and thalamic nuclei. Projections of this sort may allow frontal cortex to direct thalamic control of selective attention (“spotlight of attention,”Crick, 1984). Furthermore, other projections have been found from ventromedial frontal cortex to rostral TRN and limbic thalamic nuclei, suggesting that specific ventromedial frontal networks influence thalamic control over general alertness, as well as more focused attention (Zikopoulos & Barbas, 2006). We hypothesize that ventromedial frontal networks manifest the disruption in thalamocortical function during absence seizures that are preferentially engaged in the normal cortical control of important aspects of consciousness (Tucker et al., 2007).
Other investigators have studied absence with fMRI, with the most consistent findings showing mainly thalamic activation and simultaneous deactivation in certain cortical areas, including ventromedial frontal lobe, dorsolateral, posterior cingulate, and parietal-frontal cortex during ictal discharges (Gotman et al., 2005; Hamandi et al., 2006; Laufs et al., 2006). Investigators propose that the cortical regions deactivated in absence are the same regions activated during the “default mode” of consciousness (Raichle et al., 2001), and that suppression of the normal pattern of the default mode may explain the symptomatology of absence (Gotman et al., 2005; Laufs et al., 2006). The reports of Holmes et al. (2004) and Tucker et al. (2007), while providing evidence of engagement of ventromedial frontal cortex, do not show involvement of other cortical regions of the default mode that are deactivated during fMRI experiments. In cats, TRN connections are absent to the anterior ventral (AV) nucleus, a key thalamic structure linked to the limbic system (Steriade, 2003), and, by extension, to key cortical areas activated during the default mode. If TRN-AV connections are lacking in humans as well, and given the evidence of pathology in ventromedial frontal cortex-TRN circuits, one may speculate that suppression of default mode territories is a secondary, rather than primary, effect of absence (Tucker et al., 2007). Furthermore, if propagation of the ictal discharge is limited mainly, or exclusively, to frontal-TRN-thalamocortical circuits, without spread to limbic system and default mode networks, an explanation for the rapid recovery of memory and orientation after an absence seizure may be found in the sparing of temporal-limbic networks for memory consolidation. In contrast, in temporal lobe epilepsy, where seizures may propagate through temporal-limbic circuits critical for the consolidation of episodic memory (Nadel & Moskovitch, 1997), patients may be impaired for minutes or hours postictally (Squire et al., 1976).
Dense Array EEG and “Localization-related” Epilepsy
Studies of interictal spikes
Dense array EEG studies have examined both interictal and ictal epileptiform patterns in patients with medically refractory localization-related seizures. Interictal dense array studies may be conducted on a short-term, outpatient basis (i.e., 1–2 h recording time) and are therefore relatively easy to perform, while at the same extracting potentially useful information (Phillips et al., 2002; Michel et al., 2004; Holmes et al., 2005a). In a recent study (Holmes et al., 2005a) of eight subjects, all surgical candidates, spikes were detected with Reveal software (http://www.eeg-persyst.com), and confirmed by visual analysis. For each patient spikes were clustered into populations and each spike population was subjected to source analysis at 10 msec intervals along the time course of the spike, utilizing the linear inverse method of LAURA. Although standard visual analysis suggested that all spike populations in all patients were confined to one temporal lobe region, more complex spatiotemporal patterns of spike propagation were often observed in the dense array EEG data. In some cases, sources indeed remained confined to one temporal lobe throughout the duration of the spike. In other instances, however, propagation spread rapidly from basal temporal to adjacent lateral temporal lobe, then to orbitofrontal cortex, and finally to the opposite temporal lobe. Evidence that extratemporal regions and both temporal lobes may contribute to the sources of a single “temporal lobe” spike population may give credence to the concept that temporal lobe epilepsy is a complex, often bilateral, corticolimbic network disturbance (Ebersole, 1997; Spencer, 2002). Further research is necessary to determine the importance of these findings, including the question as to whether or not interictal spike propagation patterns recapitulate the epileptogenic network activated during clinical seizures.
Dense array long-term EEG video monitoring
LTM is necessary to realistically record seizures for most patients with localization-related epilepsy. Furthermore, EEG-video documentation of the clinical seizures is a critical component in the evaluation of the majority of patients with difficult epilepsy if surgery is considered as a potential treatment option. At the University of Washington Regional Epilepsy Center, over 40 patients to date have been monitored with dense array LTM for periods of 24–96 h, with clinical seizures recorded in over 90%. All patients have had medically refractory epilepsy, and all are potential surgical candidates. Dense array LTM studies are undertaken when the standard studies fails to provide adequate ictal localization, and are performed prior to invasive LTM. Patients who have had previous brain operations or skull defects from any cause will not good candidates for dense array EEG studies until individual head models can be constructed routinely to characterize the conductivity anomalies of the skull defects.
Determination of seizure onset
The most important task in analyzing dense array EEG-recorded data is determining the point when the seizures begin. This determination is usually more difficult in localization-related than in generalized seizures, since ictal onsets in the former may be often subtle, or contaminated by artifact, and less distinguished from the background EEG compared with the wave–spike complexes of the latter. To accomplish this goal, which ultimately is based on clinical judgment, the reviewer makes use of all the skills utilized for the visual interpretation of standard clinical EEG. At present, ictal onsets are determined primarily on the basis of pattern recognition. Following video review of the clinical event, one useful strategy is to initially review the data by “reducing” it to a standard, subsampled montage and then “expanding” the data to view the entire 256 channel array, once the approximate time of seizure onset has been found. After artifact has been identified and “bad” channels replaced, the reviewer may switch back and forth between methods of visualizing the EEG, displaying the same information with an average reference “topographic” plot, voltage map, or standard chart view. Each technique has strengths and limitations, but the aggregate of all methods will, in most cases, allow the reviewer to judge the time point when the ictus begins, and commence the process of source analysis from that time point. One frequently finds that ictal onset is heralded by oscillations in the scalp recording that are clearly different from the interictal background; searching for these patterns is critical. At times, ictal oscillations are subtle and may not be observed on standard, subsampled montages; occasionally such oscillations may be slow (less then 2 Hz), and can be distinguished from artifact. It is also helpful to identify on topographic displays the “inversion lines” (reversal of voltage potential) that assists in demarcating the onset of ictal EEG discharges. In comparing dense array with standard 19–21 channel EEG recordings, it must be emphasized that ictal EEG patterns will often emerge when the full 256 channel display is shown, when such patterns are obscure or not easily identified when visualizing the subsampled data.
Case studies of localization-related seizures
A review of several individual case studies will serve to illustrate some of the findings resulting from the analysis of the onset and propagation of localization-related seizures:
1A 23-year-old man presented with medically refractory seizures since age 14. The seizures are manifested by episodes of loss of consciousness, automatisms, and postictal confusion lasting 30–60 min or more. Standard EEG studies revealed left temporal spikes, but seizures could not be lateralized on standard LTM. Magnetic resonance imaging (MRI) is normal. Clinical seizures were recorded during dense array LTM studies. Source analysis at time of ictal onsets suggested that seizures consistently originated from the left basal temporal region, but within one second, discharges propagated to the opposite, right basal temporal region (Fig. 4). Subsequent invasive subdural LTM studies were convergent with these findings. This case study illustrates the bilateral involvement not infrequently observed in subjects with temporal lobe epilepsy.
2An 18-year-old man sustained a closed head injury at age 7. Seizures began shortly thereafter and proved to be medically intractable. The attacks are characterized by sudden loss of consciousness, vocalization, clonic jerks, and rapid postictal recovery. Standard EEG demonstrated bilateral, frontopolar spikes, with left hemispheric preponderance. The attacks could not be lateralized on the basis of standard LTM, and the MRI was normal. Dense array LTM captured clinical seizures that began in left anterior frontal regions, with rapid subsequent spread to ventromedial frontal cortex, as was found in absence (Tucker et al., 2007). This suggests that in some individuals with frontal lobe seizures, ictal propagation patterns are similar to those found in “generalized” seizures, including likely preponderant activation of corticothalamic networks.
3A 39 year-old woman has had refractory seizures since age 5. During her seizures, she vocalizes, becomes unresponsive, exhibits vigorous motor automatisms (e.g., thrashing, “bicycling”), and is confused for 15–30 min or more postictally. Standard LTM recorded clinical seizures that could not be localized. MRI suggested a region of right anterior frontal cortical dysplasia. Source analysis of seizures recorded during dense array LTM provided evidence that seizures originated from the right anterior frontal cortex, near the area of dysplasia, followed by rapid involvement of the right basal temporal region. The propagation patterns that show propagation patterns to limbic structures contrasts to that found in the preceding case, and suggests the hypothesis that some extratemporal seizures propagate primarily through corticolimbic pathways.
4A 13-year-old girl presented with medically refractory, daily complex partial seizures since age 5. Her attacks, each lasting 30–60 s, consist of the onset of confusion with orofacial and upper extremity automatisms. Standard EEG studies disclosed left occipital and left parietal spikes. Ictal onsets from standard LTM were found to be of probable left posterior quadrant onset, but were poorly localized. MRI was normal. Dense array LTM studies captured one of her habitual seizures and disclosed that the seizure originated from left posterior inferior occipital lobe. Within one sec ictal propagation to right posterior temporal-occipital cortex, and then back to left parietal cortex was found (Fig. 5). The patient subsequently underwent invasive LTM, with intracranial subdural grid electrodes placed over left posterior quadrant, and subdural strip electrodes placed bilaterally over posterior quadrants and basal temporal regions. Both ictal onsets and propagation patterns recorded from the invasive EEG studies corresponded closely to that found on the dense array EEG studies (Fig. 6). Surgery was carried out, based on the results of the invasive studies, and the patient has been seizure-free 2 years after resection (Holmes et al., 2005b). This case demonstrates convergence of the findings found on dense array LTM with direct intracranial EEG recordings and postsurgical outcome. It is illustrative of one of our research goals, which is to test the validity of dense array EEG by detecting and characterizing ictal EEG signals that cannot be elucidated with conventional EEG, and comparing the findings with invasive LTM and surgical results. We are in the early stages of this project, and it is premature to draw definite conclusions. Eight subjects to date have undergone surgical resections after preoperative dense array and invasive LTM. One subject has had a good postoperative outcome; it is too early to judge outcome for the remainder. Ten patients are awaiting invasive LTM.
What does dense array EEG teach us about epilepsy?
Analysis of ictal patterns in both generalized and localization-related seizures with dense array EEG leads to the interpretation that all seizures, including those classified as “generalized,” involve specific cortical networks. The standard categorization of epileptic seizures is largely a reflection of the inadequacy of conventional EEG analysis, where spatial resolution is limited, at best. Furthermore, examination of regions of cortical involvement at the onset and during propagation leads to the hypothesis that epileptic seizures may be considered as fundamentally corticothalamic or corticolimbic in nature. Discharges in absence involve ventromedial frontal cortex, and by inference, the TRN, with sparing of limbic circuits. Absence may be the prototype of the corticothalamic or “generalized” seizure in which corticothalmic mechanisms interact with TRN networks and therefore influence thalamocortical projections and widespread cortical regions. In contrast, the temporal lobe seizure may be the prototype of the corticolimbic seizure. As shown in the second case study, some frontal lobe seizures may also show spread to ventomedial frontal cortex, in a manner similar to absence, and may also be “corticothalamic” seizures. On the other hand, other extratemporal seizures show propagation to temporal lobe structures, as shown in the third and fourth case studies. Seizures classically considered to be localization-related may therefore exhibit patterns consistent with primary involvement of either corticothalamic or corticolimbic networks. Although the observations with dense array EEG have been important in forming this hypothesis for categorizing epilepsy syndromes, observations on psychological recovery from the seizure may also be diagnostically useful. What we describe as corticothalamic seizures (commonly considered to be “generalized”) disrupt consciousness but are followed by rapid recovery of ongoing memory for context. On the other hand, what we hypothesize to be corticolimbic seizures (usually considered to be “partial” or “localization-related”) are characterized by extended postictal confusion and gradual recovery of memory and orientation to context.
What is the potential role of dense array EEG?
In addition to suggesting new insights into anatomical mechanisms of epilepsy, dense array EEG may be useful in localizing the site of seizure onset. Ideally, this determination should be made noninvasively, but in, practice, at least 30–50% of surgical candidates will require some form of intracranial EEG evaluation, including the majority of individuals with difficult extratemporal epilepsy (Holmes, 2006). Dense array EEG may hold the promise to assist in ictal localization, when standard EEG methods fail, and may reduce the need for invasive studies. However, the present evidence is preliminary and further research is needed to establish the precise role of this technique in the presurgical evaluation. At the very least, it is probable that dense array EEG will help guide the placement of invasive electrodes, if not eliminating the need for such studies in some cases. As a corollary, the method may also eventually assist in planning the intracranial placement of novel treatment devices (Motamedi & Lesser, 2006).
Technological development in dense array EEG is anticipated in several areas. Firstly, the current forward model used in source analysis calculations, which utilizes a standard MRI or ellipsoidal multishell model, may be replaced by the individual patient's own MRI. Research is currently underway to feasibly incorporate patient-specific MRIs into source analysis algorithms. Secondly, more than 256 electrodes may eventually be recorded simultaneously from the scalp. Given that the spatial Nyquist of the human scalp EEG necessitates intersensor distances less than 10 mm (Freeman et al., 2003), as many as 600 electrodes, or more, may be required to reduce interelectrode distances to achieve the ideal dimension. Furthermore, recent studies also suggest that when optimal scalp EEG spatial resolution is obtained, examination of spatial frequency patterns even make possible the extraction of details of the cortical surface anatomy, including gyral and sulcul patterns (Freeman et al., 2006b). Other future developments are likely to include the incorporation of advances in the quantitative evaluation of the onset and propagation of ictal EEG patterns, including analysis of direct current (ultraslow) frequencies (Vanhatalo et al., 2003; Miller et al., 2007), high-frequency EEG (Worrell et al., 2004), and coherence and spatial pattern analysis (Freeman et al., 2006a). Finally, the utility of dense array EEG in source localization of cortical activity can be evaluated, and improved, through joint recordings with whole-head magnetoencephalography (Liu et al., 2002). Future research will establish the clinical utility and role of each of these newer methods in extending the information from scalp EEG recordings.
The author wishes to recognize the following collaborators: University of Washington: Kerry Baker, Shahin Hakimian, Kai Miller, John Miller, Jeffrey Ojemann, Ceon Ramon, Russell Saneto, Gagan Wig; Electrical Geodesics, Inc.: Micah Brown, K. Jeffrey Eriksen, Phan Luu, Jason Quiring, Don Tucker; University California, Berkeley: Walter Freeman; Universidad National Mayor de San Marcos, Lima, Peru: Marizabel Rozas Latorre; University of Helsinki, Helsinki, Finland: Sampsa Vanhatalo.
Conflict of interest: The author has received no financial support, consulting fees, or research funding for the work described in this manuscript. He has no conflict of interest.