Está en la página 1de 18

Epilepsia, **(*):118, 2007

doi: 10.1111/j.1528-1167.2007.01381.x

CRITICAL REVIEWS AND INVITED COMMENTARY

EEG source localization in focal epilepsy: Where are


we now?

Chris Plummer, A. Simon Harvey, and Mark Cook

Centre for Clinical Neurosciences and Neurological Research, St Vincents Hospital, Fitzroy, Victoria, Australia;
Departments of Medicine and Paediatrics, University of Melbourne, Parkville, Victoria, Australia;
and Department of Neurology, Royal Childrens Hospital, Parkville, Victoria, Australia

SUMMARY

ing some of the key studies performed in the field,


with emphasis given to clinical work published in
the last five years. In doing so, we discuss why
ESL techniques have not made an impact on routine epilepsy practice, underlining some of the current problems and controversies in the field. We
conclude by examining where ESL currently sits
alongside magnetoencephalography and combined
EEG-functional magnetic resonance imaging in the
investigation of focal epilepsy.
KEY WORDS: Source modeling, Dipole, Distributed, Electroencephalography, Magnetoencephalography, Functional magnetic resonance
imaging.

Electroencephalographic source localization (ESL)


by noninvasive means is an area of renewed interest in clinical epileptology. This has been driven by
innovations in the computer-assisted modeling of
dipolar and distributed sources for the investigation of focal epilepsy; a process fueled by the everincreasing computational power available to researchers for the analysis of scalp EEG recordings.
However, demonstration of the validity and clinical utility of these mathematically derived source
modeling techniques has struggled to keep pace.
This review evaluates the current clinical fitness"
of ESL as applied to the focal epilepsies by examin-

C ONCEPTS AND C ONTROVERSIES

to the clinical validity and utility of these mathematically complex, often nonintuitive, modeling techniques. As
such, will ESL ever realize a place in the routine workup of patients with focal epilepsy? If so, what is the current level of evidence and what additional evidence is
required for ESL to achieve this status? Has ESL been
swept aside in recent years by the newer neuroimaging
modalities of magnetoencephalography (MEG) and combined EEG-functional magnetic resonance imaging (EEGfMRI)?

In the last five years, research in the field of electroencephalographic source localization (ESL) produced more
than 150 scientific papers on computer-assisted mathematical techniques for dipolar and distributed source modeling. By comparison, less than half of this number of publications addressed the clinical validation of such techniques for the investigation of focal epilepsy. Most clinical studies featured less than 20 subjects and few were
conducted prospectively. Such an imbalance might be explained away by the relative efficiency with which ESL
simulation studies yield publishable results, particularly in
the view of the advancing computational power and data
storage capacity of the modern PC processor. However,
this explanation falls short of addressing the confusion,
even cynicism, among neurologists and neurosurgeons as

The fundamentals of ESL


As a discipline that aims to localize the sources of electric currents within the brain that give rise to recordable
potential fields at the scalp (Fig. 1), ESL is almost as old
as the science of EEG itself (Jayakar et al., 1991). Since
the days of paper analog recordings, ESL in the generic
sense has been geared up in the last few decades by
computer-assisted source modeling techniques on the back
of digital EEG technology. Computer-assisted ESL (we
now limit ESL to this context) brings with it a new set of
challenges to the same goal that faced first generation electroencephalographers; that is the noninvasive localization

Accepted October 9, 2007; Online Early publication xxxxxx


Address correspondence to Chris Plummer, Centre for Clinical Neurosciences and Neurological Research, St Vincents Hospital, 5th Floor
Daly Wing, 35 Victoria Parade, Fitzroy, Victoria, Australia 3065. E-mail:
chris.plummer@svhm.org.au
Blackwell Publishing, Inc.
!
C 2007 International League Against Epilepsy

2
C. Plummer et al.

Figure 1.
(A) Scalp recorded 19-channel EEG using Common Average Reference from a patient with BFEC. Values at the
extreme right of channel waveforms correspond to surface potential points (in microVolts) at the 24.22 msec
latency marker (midway between onset and offset interval markers, asterisked). As is customary, negative values
indicate deflections above the zero potential line. The MGFP curve for the BFEC sharp wave complex is shown.
Note the earlier, small surface negative frontal sharp wave (at onset marker, maximal at F3) and the later, larger
surface negative temporal sharp wave (at offset marker, maximal at T5). (B) Butterfly plot showing superimposed
waveforms for the 19 channels. (C) Isopotential field plots for 36-msec period following peak of early spike wave
component of same BFEC discharge. Note the polarity inversion of the dipolar field that occurs across the course
of the discharge. The left frontal field is initially surface negative (blue) and later surface positive (red), while the left
temporal field is initially surface positive (red) and later surface negative (blue). The isopotential lines demonstrate
that the later surface negative temporal field has a broader distribution than the earlier, more concentric, negative
frontal field. Abbreviations: BFEC (Benign Focal Epilepsy of Childhood), MGFP (Mean Global Field Power).
C ILAE
Epilepsia !
of epileptogenic networks in the patient presenting with
epilepsy.
Two fundamental problems exist in the practice of
ESLforward and inverse. The forward problem is solved
by specifying a set of conditions (compartments, surfaces, conductivities) for the head model, also referred to
as the volume conductor or forward model. The forward
model by analogy is the stage on which the source, or
source network, performs, its projected (lead) field passing through modeled compartments and tissue interfaces
to reach the recording electrodes. It is the set of conditions specified to the forward problem that distinguishes
one forward model from another. Forward models range
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

from simple (a single spherical shell models the brain surface) to complex (a four-layered realistic model, its compartments segmented from the patients MRI scan, models the brain, cerebrospinal fluid, skull, and scalp surfaces). Spherical shell and realistic models are the two versions of forward modeling used in ESL today (Fig. 2).
The former vary in complexity from single shell to multiple (two, three, or four) overlapping shell models, and
the latter are sub-divided into boundary and finite element
method (BEM, FEM) models. We touch on the application
of these models in the next section. For a specific electrical source, the forward model will enable the computation of a specific potential field at its surface (Wilson &

3
EEG Source Localization in Focal Epilepsy

Figure 2.
The evolution of forward modeling in ESL. Spherical shell models range in complexity from one to four overlapping
shell surfaces to model the head as a volume conductor (single shell- brain, 2 shell- brain, skull, 3 shell- brain, skull, skin,
4 shell- brain, CSF, skull, skin). Realistic head models (BEM and FEM) are so called because they better approximate
the shape of the human brain than do shell models. This is particularly the case at the brains deeper, inferior surfaces,
as illustrated here with a 4-shell model projected over a digitally reconstructed cortex from an averaged MRI scan.
The BEM models are composed of overlapping, two-dimensional, triangulated mesh layers (or boundaries), each
layer having been computer generated from segmented T 1 -weighted MRI surfaces (scalp, skull, CSF, and cortex).
Different compartments are usually given different conductivity values, but conductivity within each compartment is
assumed to be isotropic and homogenous. In contrast, the FEM models are composed of multiple, three-dimensional,
solid tetrahedra, a property that allows conductivity values to vary within each compartment. This means that tissue
anisotropy can be factored into algorithms that solve the forward problem. As distinct from the 4-shell model, note
how well the FEM captures the shape of the brain in the modeling of its innermost compartment. Abbreviations: BEM
(Boundary Element Method), CSF (cerebrospinal fluid), FEM (Finite Element Method), MRI (Magnetic Resonance
Image).
C ILAE
Epilepsia !
Bayley, 1950). Thus, the forward problem will give a
unique solution.
The inverse problem, by contrast, has no unique solution. That is, an infinite number of source permutations
can, in theory, explain a specific potential field recorded
at the surface (Helmholtz, 1853). This is the problem the
electroencephalographer attempts to solve in routine clinical practice, traditionally with a minds eye rendering of
candidate sources drawn from the various EEG montage
digital displays. In practice, the experienced electroencephalographer constrains the infinite possible solutions
to the inverse problem by applying their working knowledge of epileptogenesis in the focal epilepsy syndromes to

the patients clinical picture, supplemented perhaps by information from anatomical or functional imaging studies.
Similarly in ESL, the inverse problem is made soluble by
the incorporation of mathematical constraints into inverse
modeling algorithms.
Just as volume compartment and boundary conditions distinguish one forward model from the next, constraint conditions distinguish one inverse model from
the next. The two major inverse modeling approaches
are the dipolar and distributed modeling methods. The
mathematics of inverse modeling can be quite complex
and it is not the purpose of this paper to summarize
the algebraic pros and cons of the several dipolar and
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

4
C. Plummer et al.

Figure 3.
(A) ESL example of a Moving dipole inverse model used with an FEM forward model based on the same BFEC sharp
wave complex depicted in Figure 1. Each dipole is fitted to a single time point at successive 4-msec intervals across
that part of the interictal discharge shown in Figures 1A, 1B. Relative dipole size is in proportion to dipole strength
or amplitude. Note that each color-coded dipole symbol is made up of a spherical end directed toward the surface
negative cortex and a straightened tip directed toward the surface positive cortex. Hence, the earlier, smaller surface
negative frontal sharp wave seen in the BFEC discharge is represented by the smaller, darker green dipoles, while the
larger, lighter green dipoles represent the upswing phase of the larger, later surface negative temporal sharp wave.
Also note that an example of a confidence ellipsoid is shown attached to one of the larger dipoles (asterisked). It
is worth pointing out that different types of dipole symbols are used in the ESL literature, but investigators do not
routinely spell out surface negative and surface positive aspects of such dipole symbols. (B) LORETA distributed inverse
model used with a BEM forward model. Note the relatively diffuse distribution of the current density map spanning
the left central sulcal region. Broad, blurred ESL solutions are quite typical for the LORETA method. Brighter signal
intensities indicate higher current density values (scaled as microamperes per square millimeter). Orthogonal axes
are shown (+x left, +y posterior, +z upward). Appreciate that the single time point chosen for display (24.22
msec) approximates the halfway point of the upswing phase of the surface negative discharge, a point which, on current
evidence, appears to most reliably reflect the state of the corresponding intracranial potential field. Abbreviations:
BEM (boundary element method), BFEC (Benign Focal Epilepsy of Childhood), ESL (EEG source localization), FEM
(finite element method), LORETA (Low Resolution Electromagnetic Tomography).
C ILAE
Epilepsia !

distributed models available in ESLfor reviews see (Darvas et al., 2004; Michel et al., 2004a). What should
be stressed though is that dipolar methods are overdetermined (Fuchs et al., 1999) in the sense that the investigator preselects one, two, or three (rarely more) dipoles
to apply to the inverse algorithm in question. This means
there are far more data sampling points (viz. electrodes)
than there are dipole parameters in determining the ESL
solution.
Two of the most commonly used inverse algorithms in
ESL are the moving and rotating dipole methods. The moving algorithm constrains the dipole to instants in time (successive 4 msec time instants in a 256 Hz recording are usually solved); but frees it in space to assume a location, orientation, and strength to best explain the measured EEG
data at each time instant (Fig. 3A). The rotating algorithm
constrains the dipole to a location in space; but frees it to
assume an orientation and strength to explain the variance
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

in the measured data across any time interval (Fuchs et al.,


2004a).
In contrast, distributed methods are under-determined
(Fuchs et al., 1999). That is, there are far fewer sampling
points than possible ESL solutions. This is because, unlike dipole modeling strategies, no assumption is made
on the number of dipoles used to solve the inverse problem. Instead, the working premise is that multiple sources
may be simultaneously active across multiple locations at
a given instant in time. The predefined solution space (be
it the whole brain volume or just the cortical volume) is
split into multiple points, each point representing a minidipole, fixed in space but free to assume any orientation
and strength. Due to the enormous number of permutations that stem from such mini-dipole networks, all offering a theoretically plausible explanation for the measured
EEG signal, postprocessing constraints need to be applied
to achieve a unique ESL solution.

5
EEG Source Localization in Focal Epilepsy
Low-resolution
electromagnetic
tomography
(LORETA) is one of the more familiar distributed
modeling algorithms used in ESL (Fig. 3B). LORETA
applies a modeling constraint based on the idea that
neighboring neuronal populations are more likely (than
nonneighboring ones) to undergo synchronous depolarization during a spontaneous discharge or an evoked response
(Pascual-Marqui et al., 1994). LORETA modeling tends to
generate broad, smoothed ESL solutions as neighborhood
sources are model term conditioned to assume similar
strengths. As discussed later, such assumptions may not
always be in touch with electrophysiological reality.
It is often misunderstood that the inverse and forward
problems are interdependent. They are only interdependent
in the sense that both are required to generate an ESL solution. However, the mathematical algorithms specified to
each problem are independently set (Scherg et al., 1999).
This is important to appreciate because any ESL solution
can be reached when any inverse model is coupled with
any forward model. The two major challenges then, lie in
(a) deciding on which of the available forward and inverse
models are the most appropriate to apply, and (b) translating the theoretical impact governed by the choice of a
particular forwardinverse modeling set-up into terms that
define the clinical impact of such a choice on the patients
diagnosis and management. Scant attention has been given
to the latter in the ESL literature to date. Most clinical studies preselect a particular forwardinverse modeling combination (shell or realistic-dipolar or distributed) and test its
performance in a particular clinical setting either directly,
using simultaneously acquired intracranial EEG, or more
commonly indirectly, using estimates of concordance with
other functional and anatomical studies.
On the forward problem and its problems
The relative disconnect at the technicalclinical interface of ESL research is exemplified by the forward problem. The digital reconstruction of realistic forward models that predict the impact of volume conduction on the
generation of scalp potentials remains a central theme in
biophysics EEG research. But on this point, Niedermeyer
reminds us of Jaspers observation 40 years ago that propagation along conducting pathways represents the most important mechanism of signal spread (especially as regards
epileptiform signals) (Jasper, 1969), himself warning that
excessive emphasis on volume conduction and total reliance on biophysics are not the answer to EEG signal
analysis (Niedermeyer, 2005). There is little doubt that
the modeling of basal source activity, as in temporal lobe
epilepsy, is optimized with the use of realistic head models that more accurately delineate the nonspherical, inferior aspects of the brain compared to overlapping shell
models (Ebersole, 1997a). The latter give dipole location
errors of up to 30 mm in the rostral directionspherical
shell models commonly mislocalize known mesial tempo-

ral lobe source activity to the frontal lobe (Cuffin, 1996;


Roth et al., 1997). Errors in dipole orientation were also
noted to increase when spherical models were used in
place of realistic models in a more recent simulation study
(Crouzeix et al., 1999). While these observations are important, it should be appreciated that the effects of signal
propagation (primarily via cortico-cortical pathways) on
scalp EEG voltage topography are not factored into equations designed to solve the forward problem.
It is probably under-emphasized that in terms of presentday clinical utility, the property that differentiates spherical from realistic head models simply relates to model
shape, rather than to any more advanced feature (Fuchs et
al., 2002). Spherical models conform to frontal and parietal brain convexities reasonably well but are found wanting when it comes to the modeling of infero-occipital,
infero- and mesial temporal, and orbitofrontal brain surfaces (Ebersole, 2003a), regions that commonly play
host to the epileptogenic zone in focal epilepsy. Research
efforts aimed at further advancing realistic models, such
that they are brought closer to reality, remain hamstrung
by three factors.
The first is the extra computational demand that complicates the integration of tissue anisotropy parameters
into the forward modeling algorithm. While finite element
methods aim to satisfy the more physiological anisotropic,
heterogeneous conduction that takes place within tissue
compartments and across tissue boundaries (Fuchs et al.,
2007); in practice, most realistic models (viz. boundary element methods) assume homogenous, isotropic conduction
within compartments and limit conductivity variability to
surface boundaries alone.
The second factor is the absence of agreed-upon tissue
conductivity values for the various tissue compartments in
the human head. For example, the wide range of values
published for the skull-to-brain conductivity ratio in humans is at least partly due to inherent discrepancies between in vitro and in vivo based findings (Nunez & Srinivasan, 2006a). Clinical studies often apply conductivity
values for the brain, skull, and scalp that are based on an
in vitro study published 40 years ago (Geddes & Baker,
1967). In any case, a more recent in vivo study suggests
that the inter-individual variability of these properties may
limit the generalizability of tissue conductivity values for
realistic models (Ha et al., 2003). Moreover, while it has
been theorized that the brainskull interface dampens the
voltage of a dipole point to one-eighth of its strength, it
should be kept in mind that skull conductivity actually
improves with increasing skull thickness due to the accompanying increase in the ratio of cancellous to cortical
bonemarrow conductivity being higher than that of cortical bone (Nunez & Srinivasan, 2006a). Thus, the individualization of forward modeling would expectedly depend
on regional nuances in individual skull thickness and cranial contouring relative to the brain, the latter influencing
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

6
C. Plummer et al.
electrode distance and electrode orientation relative to the
cortical surface (Binnie et al., 1982).
The third problem that needs to be resolved is the clinical relevance of incorporating brain tissue anisotropy into
forward modeling calculations via finite element methods. Average white matter resistivity is approximately double that of gray matter, largely by virtue of the multidirectional nature of white matter fiber tracts (Nunez &
Srinivasan, 2006a). While diffusion tensor imaging (DTI)
techniques are beginning to avail quantification of white
matter anisotropy (Mori & van Zijl, 2002), the relative clinical impact of such a parameter on volume conduction (vs
brain-skull and scalp-air interface effects) remains unqualified. Although speculative, anisotropic modeling of brain
tissue may offer some much needed insight into the biophysical properties of interictal and ictal source propagation occurring across cortical regions. The immediate relevance of this becomes apparent when one considers that
approximately 98% of pyramidal cell input in humans is
via cortico-cortical connections (Braitenberg, 1978).
On the inverse problem and its misconceptions
Just as forward models vary in complexity from single
shell models to multicompartment realistic models, inverse
dipolar models range from single fixed dipoles fitted to single time points (as with fixed and moving dipoles) to multiple dipoles that overlap in three-dimensional space across
time intervals (as with rotating and regional spatiotemporal dipoles). Likewise, distributed models can be broadly
classified into those that use linear mathematics (as with
the so called minimum norm, depth-weighted minimum
norm, and LORETA models) and those that use nonlinear
mathematics (as with L1 norm methods). However, rather
than presenting a didactic discussion on the variety of inverse algorithms, both dipolar and distributed, available in
ESL today (some of which we discuss further in the next
section), we will address some of the key concepts in inverse modeling as a series of common misconceptions.
(A) Dipole models provide point-like anatomical
solutions for spike and seizure localization
It is tempting to interpret dipoles as point sources of interictal or ictal activity when they are pitched, as is often
done in publications, over a coregistered image from the
patients MRI scan. This is especially so when ESL studies not uncommonly infer that a dipoles x, y, z location
in space is the sine qua non of a dipole models accuracy.
This misconception is understandable given the millimeter
margins of error that are often cited for dipole positions
in simulation studies, particularly in the biophysics literature. As Ebersole has repeatedly emphasized, a dipole is
not a discrete anatomical construct but a theoretical concept on which the modeling of relatively large segments
of synchronously discharging cortex is based (Ebersole &
Hawes-Ebersole, 2007). The threshold cortical area for a
spike to be seen by the scalp electrodes is 10 cm2 (Tao
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

et al., 2005). The orientation of the dipole is just as, if


not more, informative of the behavior of a putative source.
The state of a dipoles orientation within milliseconds of
spike or seizure onset can shed light on patterns of corticocortical propagation, a phenomenon not easily appreciated
from the traditional visual inspection of the EEG waveform
alone (Ebersole, 1994).
(B) The dipole marks the center of mass of the source
This is an oft-quoted phrase in the ESL literature. Unfortunately, as a definition it is somewhat vague and, as Gotman points out, there is actually no proof for it despite its
longevity as a concept (Gotman, 2003). Rather than visualizing a dipole as the center of a mass of equivalent current,
simultaneous surface-depth recordings suggest that, for a
given time point or time interval, the dipole most reliably
models the maximum potential of a source whose intracerebral field is often quite extensive (Merlet & Gotman, 1999)
and whose strength, location, and orientation can change
quite rapidly across the time course of an epileptiform discharge.
(C) As artificial concepts, dipole and distributed models
carry little electrophysiological relevance
Electroencephalograph literally means electrical brain
picture electro-encephalo-gramma being the Greek
roots (Knott, 1985). Our present understanding of EEG signal generation is actually based on the electrophysiological
theory that the EEG waveform is the product of a myriad
of dipoles, or dipolar configurations, that flux in polarity
within the cortical space (Brazier, 1949; Gloor, 1985; Ebersole, 2003b). The electromotive force behind dipolar field
generation is the resting membrane potential of the pyramidal neuron. Following its excitation at the postsynaptic membrane, the pyramidal cell experiences a progressive
wave of depolarization along the length of the axon. Passive loops of extracellular current are set up to complete
the local circuit (Buzsaki et al., 2003). It is the linear summation of the extracellular components of these pyramidal
cell microcircuits, positive and negative, that configures the
source and its projection to the scalp as a recordable potential field. It is useful to visualize this extended source activity through the conceptual lens of Gloors solid angle theory (Gloor, 1985). Perhaps the most common error in EEG
reading is to see the electrode that registers the peak voltage of an epileptiform discharge as the one that lies closest
to the source (Ebersole, 2000). Gloors theory reminds us
that it is the cortical configuration (area and orientation)
of the source in relation to the recording electrode, rather
than the source-to-electrode distance, that determines EEG
surface polarity. With this is mind, it can be appreciated
that both surface polarity maxima (positive and negative)
of the potential field carry useful localizing information
(Fig. 1C). For instance, it is typically the contralateral
surface-positive, and not the ipsilateral surface-negative,
potential field maximum that better delineates the origin

7
EEG Source Localization in Focal Epilepsy
and propagation of ictal or interictal source activity in
mesial versus lateral temporal lobe epilepsy (Ebersole &
Wade, 1990).
In electrophysiological terms, it is important to keep the
spatial limitations of dipole modeling in perspective. One
cubic millimeter of human neocortex contains around 105
neurons and 109 synapses. One scalp electrode is estimated
to record the synchronized and aligned space-averaged potentials of around 108 to 109 neurons (Nunez & Srinivasan,
2006a). One dipole models upwards of around 10 cm2 of
cortical tissue. This is why it is important not to regard
dipole solutions as point-like millimeter (let alone submillimeter) indices of abnormally discharging cortex. By
the same token, such a logarithmic leap in spatial dimension has led Nunez to propose that, should there be a major shift in our future understanding of epileptogenesis at
the micro-scalar level (cellular and subcellular), the electrophysiological relevance of dipole modeling theory at the
macro-scalar level (lobar and sublobar) will likely remain
intact (Nunez & Srinivasan, 2006a). It is rather the next
tierthe application of macro-scalar theory, in the form
of ESL, to routine epilepsy patient work-upthat needs
more rigorous proof of concept at this point in time.
(D) The surface negative peak of the highest amplitude
spike from the scalp EEG recording gives the most
reliable ESL result
This is both partly true and false. It is technically true
because dipole and distributed modeling solutions are most
stable when the signal to noise ratio (SNR) is highest, as is
usually the case at the spike peak. The stability of an ESL
solution, or more specifically, the parameters that define it
(location, orientation, strength), relates to its reproducibility and, pending the suitability of the forwardinverse modeling set-up, to its capacity to explain the signal variance
at the scalp electrodes. Along these lines, a recently introduced strategy to help quantify the probability of an ESL
solution is the confidence ellipsoid (CE) volume calculation (Fuchs et al., 2004b). Dipoles fitted with CE volumes
are, by definition, free to roam within the confines of the
ellipsoid space without its inverse fit parameters impacting on the forward fit solution beyond the level of noise
attached to the solution subspace (Fig.3A). In other words,
the smaller the CE volume, the greater is the probability
that the dipole resides at the fit location for a given time
point or time interval. An inverse relationship between the
CE volume and the SNR has been demonstrated in both
simulation (Fuchs et al., 2004b) and clinical (Plummer et
al., 2007) studies. Thus, small CE volumes tend to occur in
the vicinity of the spikes peak where the SNR is typically
higher.
The above tenet is, however, misleading in terms of the
probability that the ESL result actually models the original
interictal or ictal source. It has become increasingly recognized that ESL results based on spike peak activity, an ap-

proach that is still seen in clinical studies, should be interpreted with caution. This is because simultaneous surfacedepth recordings reveal, perhaps not surprisingly, that it is
the earlier component of the epileptiform discharge at the
scalp, which most closely matches the location and field of
the source as suggested by the corresponding intracranial
EEG activity (Fig. 1A, 1B). At the scalp recorded spike
peak, the signal is often well removed from the original
source due to the effects of cortico-cortical propagation.
The problem then lies with the accurate modeling of earlier phase interictal or ictal activity when the signal is often
buried in noise. Scherg has emphasized that source activity onset is best demarcated with a higher low-filter setting, recommending a 210 Hz frequency threshold range
instead of the more traditional 0.51 Hz cut-off (Scherg
et al., 1999). The effect is to minimize the contribution of
slower frequencies that are less likely to figure in the earliest source activity. In a similar vein, he stresses the importance of using a forward noise filter for ESL, rather than a
zero-phase shift filter, as the latter tends to artificially blur
signal onset and offset.
Assuming technically satisfactory EEG signal acquisition, the SNR is commonly optimized by averaging single events. However, averaging carries the inherent risk
of mislocalizing single events if the latter are not truly
monomorphic (Braga et al., 2002; Chitoku et al., 2003).
Various methods, such as phase coherence and global field
power correlation (Lehmann, 1987), have been used to help
pool identical discharges for averaging purposes. While
averaging can improve the localizability of the earlier
components of focal epileptiform discharges (to the point
where it tends to be done routinely in research), few studies have rigorously examined the clinical impact of single
versus averaged event selection in dipolar and distributed
modeling.
(E) ESL is too cumbersome to perform. Too many
electrodes, too much computer knowledge, too many
difficulties with image coregistration, and too many time
demands make it impractical for routine use in the clinical
setting
There remains no fixed agreement on the minimum
number of scalp electrodes required for clinically useful
ESL in focal epilepsy. While high-density electrode arrays can improve the spatial resolution of surface EEG
signal topography, and thus facilitate the task of distinguishing source origin from source propagation, there is
the penalty of having to measure and fix hundreds of electrodes to the scalp. On theoretical estimates, the minimal number of scalp electrodes required for optimal EEG
spatial resolution, the so-called Nyquist criterion, lies between 100 and 200 (Gevins, 1993; Srinivasan et al., 1996).
Electrode caps are not an ideal answer to this problem as
electrode-scalp contacts can be unreliable and their use
for long-term EEG monitoring is impractical. The loading
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

8
C. Plummer et al.
of scalp electrodes over putative cortical foci, as with the
use of inferior temporal arrays in temporal lobe epilepsy,
can provide clinically valid ESL results based on intracranial EEG localization (Ebersole, 2003a). Dipole simulation work has also demonstrated that nonuniform sampling
with 1010 scalp positions in the region of interest and
1020 positions elsewhere can provide reliable estimates
of source characteristics (Benar & Gotman, 2001). A more
recent study in a group of 14 patients with refractory focal epilepsy and Engel class 1 surgical outcomes showed
that ESL accuracy, indexed by the distance from the nearest surgical margin to the location of a single fit inverse
model, improved by around 2 cm from a 31 to a 63 electrode set-up, with little change from a 63 to a 123 electrode
set-up (Lantz et al., 2003a). While source orientation was
not considered and single-spike peaks were modeled, this
is, perhaps surprisingly, the first study to have systematically examined this issue in a well-defined patient group.
The question of optimum scalp electrode number for ESL
may be settled by default with the future development of
quicker methods that reliably fix high-density electrode arrays to the scalp. Until then, more studies in the manner of
Lantz et al. are needed.
Coregistration problems are more readily overcome with
the recent availability of MRI compatible electrodes and
the ability to perform less artifact-laden EEG recordings
in the MRI scanner. Newly developed coregistration methods, based on the use of mutual three-dimensional virtual landmarks from patient MRI datasets, show promise
(Fuchs et al., 2007) but await clinical validation.
The computer technology, on which current-generation
ESL relies, has become more accessible to clinicians in the
last five years due to improvements in the user interface of
software operating systems.
(F) Distributed models display sources as current density
field maps that are closer to reality than dipole
modeled sources
This, understandably, is not an uncommon misconception. The distributed-ness of a distributed algorithms solution is generally not the direct representation of the potential field of the actual source. This only holds if the distributed model is perfectly correct, which is virtually never
the case. In fact, much of the apparent field effect results from the distributed algorithms inexactness in modeling the source. Some of this modeling error can be limited
by preconstraining the ESL solution to anatomically meaningful boundaries, such as the cortex, a benefit carried by
distributed over dipolar modeling methods (Wagner et al.,
2001). However, by virtue of their under-determined nature, distributed algorithms are computationally more demanding such that, for most present applications, ESL solutions can only be calculated for time instants. This means
that it is difficult to appreciate the relative timing of overlapping source components contributing to the modeled
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

cortical activity across the early spike interval (Scherg et


al., 1999). Also, for distributed ESL solutions to be sufficiently spatially resolved, current density thresholds are set
which, not unlike fMRI signal thresholds, are typically arbitrary. Hence, the more established anatomical pathways
of interictal and ictal discharge propagation are not factored into current density threshold settings.

C LINICAL S TUDIES IN
D IPOLAR AND
D ISTRIBUTED ESL
Despite the clinically based research efforts in ESL by
Ebersole and others over the last two decades, Krauss and
Webber still have it that digital EEG has not significantly
expanded the clinical role of EEG, with the possible exceptions of ambulatory monitoring EEG and OR/ICU EEG
(Krauss & Webber, 2005). While their premise that an expanded clinical role for digital EEG may depend partially
on validating advanced analysis techniques, e.g. modeling
seizure sources, seems reasonable enough, does their implicit observation on the clinical worth of ESL still hold,
particularly in light of the work carried out in the field in
the last few years? What recent progress has been made
toward the clinical validation of ESL? We explore some
fundamental questions that warrant closer scrutiny if ESL
is to assume a clinical role in routine epilepsy practice.
Which part of the spike should be modeled?
Lantz and colleagues have carefully examined this issue
(Lantz et al., 2003b). They wondered how stable the scalp
EEG field was from spike onset to spike peak. They based
their observations on the spike-averaged recordings of 16
patients with symptomatic focal epilepsy. All had an Engel class 1 surgical outcome. Using a novel spatiotemporal cluster analysis technique, they saw, on average, three
different voltage field maps during the rising phase of the
scalp-recorded spike per patient (range one to five). When
ESL was performed on these different voltage maps, the
source model location coincided with the MRI lesion location for all patients within a fairly narrow time window
across the upswing phase of the spikearound the halfway
point. Either side of this point, the authors argued that ESL
results were contaminated by noise (toward spike onset)
and by propagation effects (toward spike peak). It should
be noted that 125 electrodes were used in the study, so the
application of a half-way point rule to ESL when fewer
electrodes are employed, as is commonly the case, is not
entirely clear. Also, because the inverse model used was
a single fit applied to a combined dipolar-distributed algorithm, EPIFOCUS (Grave de Peralta Melendez et al.,
2001; Lantz et al., 2001) for each field map, spatiotemporal
relationships between successive, independent time-point
fits cannot be fully resolved. Finally, as all spikes were averaged, the generalizability of the results to single-spike

9
EEG Source Localization in Focal Epilepsy
modeling is uncertain. Still, the study is the first to systematically quantify the increasingly appreciated concept that
the surface voltage topography, and by extension the ESL
result, shifts on a scale of tens of milliseconds across the
earliest phase of spike interval.
How well does ESL corroborate epilepsy surgery
findings?
The largest prospective study to date on this topic (Boon
et al., 2002) looked at the contribution of spatiotemporal
dipole modeling to the clinical decision making process
in 100 presurgical patients with refractory focal epilepsy.
Most cases were lesional (83%), with the largest subgroup
having unilateral hippocampal sclerosis (53%). Scalp ictal
EEG recordings from a 27-electrode set-up (1020 positions plus three inferior temporal electrode pairs) were analyzed. From the 93 patients who recorded ictal EEG phenomena, 62 patients could not undergo ESL analysis due
to excessive artifact contamination. Of the remaining 31
patients, it was concluded that ESL influenced the clinical
interpretation in 14 cases, usually by confirming the incongruence between structural abnormality and ictal EEG
abnormality (10 patients) and leading to the decision not
to proceed with invasive EEG recording and further surgical resection.
Unfortunately, the study contains several methodological deficiencies. It was not blinded, it gave no information on postsurgical outcome, and it performed zero-phase
shift filtering on the EEG raw data. The spatiotemporal
modeling was actually limited to the use of a single (regional) dipole, thus making it difficult to disentangle interlobar or interhemispheric propagation effects from effects
potentially attributable to multiple independent sources on
ESL outcome. Also, ESL results were strictly categorized
as type 1 (vertical) and type 2 (radial) dipoles, based on the
earlier observations of Ebersole and Wade, who equated
the type 1 dipole with a mesiobasal temporal lobe source,
and the type 2 dipole with a lateral temporal lobe source
(Ebersole and Wade, 1990). This classification was subsequently seen as an oversimplification by Ebersole himself,
recognizing the inter-changeability of type 1 and 2 dipolar patterns in both forms of TLE (Ebersole, 2000), largely
by virtue of discharge propagation effects occurring early
in the interictus, and even earlier in the ictus. The clinical
immediacy of this problem was reemphasized by a recent
depth electrode study that showed that postsurgical success in medically refractory TLE relies heavily on the spatial resolution of the ictal onset zone on a sublobar scale
(Chabardes et al., 2005). Lastly, the authors quoted an 8 h
time cost for the analysis from start to finish, an experience
that contradicts recent findings on the relative clinical utility of dipole modeling in focal epilepsy (Plummer et al.,
2007).
In the largest prospective interictal ESL study to date
(Michel et al., 2004b), a heterogeneous group of 44

epilepsy surgery candidates undertook a supplementary


128 channel surface recording for the purpose of single
source dipolar-distributed modeling (EPIFOCUS). Of the
32 patients who had an identifiable focus, seven of whom
underwent invasive recording in addition to the routine
presurgical work up, all but two patients had concordant
ESL findings at a lobar level. From a subgroup of 24
patients who underwent surgery (17 temporal, seven extratemporal), 18 had an ESL maximum that fell within
the border of the nearest resection margin (three temporal,
three extratemporal were nonconcordant). An Engel class 1
outcome was shared by 16 of the 18 cases (mean follow up
19 months, range: 733 months). Interestingly, two of the
nonconcordant extratemporal cases were mirror localized
to the contralateral hemisphere, their respective presurgical
MRI lesions sitting close to the parieto-occipital midline.
Although the intracranial and 128 channel recordings were
not performed simultaneously, the investigators quoted a
high level of agreement between the intracranially directed
interictal and ictal localization and the high-density surface
electrode-directed ESL result (five of seven cases).
What is especially striking about this study is the degree
of accuracy achieved for the localization despite the fact
that each patients MRI brain was morphed to fit a threeshell sphere for the forward model set-up, with standardized electrode positions prefitted to the outermost shell.
ESL results were constrained to the cortical gray matter
and based on the midway point of the averaged spikes upswing phase. While the results are encouraging, and each
ESL analysis was performed in a timely manner, there is
the concern that the investigators were not blinded to the
patients para-clinical data during the source fitting procedure. Comparative results on normal (mock) MRI data
would have further strengthened the case for the robustness of their source localization technique. Also, as the
investigators do point out, resection boundaries are variably wider than lesion boundaries and so measurement bias
may have inflated the accuracy of their ESL results. Rather
than countering this point by stressing the ESL concordance for scalp and intracranial recordings in the few patients who had both performed, a breakdown of the distance from nearest resection boundary to nearest lesion
boundary in each of the surgical cases may have been more
informative.
The same research group (Sperli et al., 2006) more
recently examined ESL accuracy in a pediatric epilepsy
surgical cohort (13 temporal, 17 extratemporal). Interictal EEG recordings were acquired using 1929 scalp electrodes. A distributed inverse model was applied in this case
(depth-weighted minimum norm, MN). The MN algorithm
favors current density solutions that explain surface electromagnetic fields with the least net strength per time point
(Hamalainen & Ilmoniemi, 1994). MN solutions therefore
typically localize to the superficial cortex and a mathematical depth weighting term is often applied to counteract
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

10
C. Plummer et al.
this tendency (Michel et al., 2004a). Presumably as a result of the blurred, diffuse nature of the MN-based solution,
ESL accuracy was determined in this study by the degree
of overlap (arbitrary 50% minimum) between the resection
boundary (defined as the epileptogenic region) and a statistically deconstructed depth-weighted MN map (earliest
vowel-wise activity p < 0.0001 vs. background), which
the authors defined as the region of discharge onset.
The results were encouraging with a 90% concordance between ESL location and nearest resection border and an
87% postoperative seizure freedom rate (mean follow up
13 months, range: 224 months). The authors argued that
the three mislocalized cases (all temporal) stemmed from
the inadequate sampling of the inferior temporal region, a
premise that was supported by their corrected modeling
of two of these cases when ESL was repeated with a highdensity electrode set-up (128 channels).
As with their previous paper, there was no blinding in
this study. In a majority of the cases (19/30), no postoperative MRI was available and the epileptogenic region was
mapped via an interpretation of the surgical notes.
This study draws out some of the inherent difficulties associated with ESL based on current-generation distributed
modeling methods. With solutions that are as visually seductive as functional MRI (Ebersole, 1997a), it should be
appreciated that distributed models are based on complex
sets of mathematical assumptions that are yet to achieve
clinical validation. The use of a statistical discriminator
in the present study might be seen as a step in the right
direction in this validation process. While the present results need to be reproduced by other investigators, the level
of accuracy obtained with a 1020 electrode set-up does
support a potential role for ESL in the routine clinical
setting.
Finally, it should be stressed that studies using lesion
or resection margins to validate ESL results, even in Engel class 1 cohorts, should be interpreted cautiously. The
relationship between the epileptogenic zone and the putative epileptogenic lesion is far from directfor review see
Rosenow & Luders (2001).
How well does scalp ESL model interictal and ictal
onset as defined by intracranial recordings?
Zumsteg and colleagues also performed statistical postprocessing of a distributed algorithm (LORETA) in a retrospective study of 15 patients with symptomatic mesial TLE
(MTLE) (Zumsteg et al., 2005). Unlike the study by Sperli
and colleagues (Sperli et al., 2006), they used nonparametric mapping (SNPM), thereby avoiding an assumption that
the raw ESL results necessarily conformed to a Gaussian
statistical distribution. They compared the interictal localization suggested by recordings from foramen ovale (FO)
electrodes, which look directly at the hippocampus, with
both the raw and SNPM LORETA results derived from the
simultaneous scalp EEG recording (23 electrode set-up).
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

From 19 local field patterns seen by the FO electrodes


(11 patterns excluded), 14 could be localized by scalp ESL
based on the rising phase of the spike-averaged waveform.
Raw LORETA maps typically showed basolateral temporal activation, while the corresponding SNPM LORETA
images showed more discrete, mesially placed activation
that correlated well with the local FO field. The decision
to exclude 11 patterns from the LORETA analysis due to
the suspected nonfocality of the source (based on the local
FO field) seems unusual. One of the benefits of distributed
over dipolar modeling is that the former is generally better
equipped to display extended source configurations. Also,
while the authors tabulated Engel class surgical outcomes
at one year for each patient (most were class 1), they did
not indicate which FO pattern belonged to which patient.
In a follow-up study (Zumsteg et al., 2006), the investigators used the same patients and the same SNPM
LORETA technique to explore the nature of spike propagation in MTLE. However, all 30 FO field patterns were
included in this study (19 mesial from the earlier study,
and 11 lateral). Based on the SNPM LORETA activation sequence, signal propagation was evident in 16 patterns, occurred in either direction (mesial to lateral, lateral
to mesial), preceded the spike peak, and was not associated
with Engel class outcome.
While the authors noted several limitations in the retrospective study design (reliance on FO electrodes to capture
large propagating fields, use of a three-shell forward model
coregistered with a generic brain MRI, and use of only two
supplementary inferior temporal electrode pairs), they regarded it as the first attempt to examine the accuracy of
a distributed inverse model using simultaneously acquired
scalp and intracranial EEG recordings.
A different group of investigators (Nayak et al., 2004)
used dipole modeling to help characterize the relationship between the FO recorded field and the corresponding
scalp EEG field in a retrospective study of 20 patients with
MTLE. Several important findings came from their meticulous analysis of over 4,000 FO spikes. Only 9% of FO
spikes were identified de novo at the scalp. Otherwise,
either scalp signal averaging (60%) or FO spike correlation (13%) was needed to confidently identify the corresponding lower voltage scalp spikes (no scalp signal was
seen despite such EEG postprocessing in the remainder,
18%). Interestingly, de novo scalp spikes were associated
with a shallower FO field gradient and were seen up to 2
msec later than the small scalp spikes. The latter were associated with a wider scalp field with around 20% of these
peaking in amplitude at the contralateral scalp. Dipole localization placed de novo spikes at the retro-orbital region
and smaller spikes at the mesial temporal region, but there
dipole orientation was more haphazard.
The authors reasoned that the de novo spikes (100+ microV) were the product of summated propagation from the
deeper mesial source configuration and that the smaller

11
EEG Source Localization in Focal Epilepsy
spikes (2040 microV) were the product of volume conduction. The implication here perhaps is that dedicated
identification of low-amplitude spikes in MTLE can provide very useful ESL information that is potentially more
physiologically plausible given the fact that dipole modeling is fundamentally reliant on the principles of volume
conduction. A criticism of this aspect of the study however, was the use of a single fixed dipole and a single shell
model. While acknowledging that this simplified modeling set-up was not designed to interrogate source behavior
at the neurophysiological level, the authors did nonetheless go on to discuss their ESL findings at this level. For
instance, it was argued that the retro-orbital dipole localization was likely the result of distortion of the scalp EEG
field with the preferential propagation of current through
the superior orbital fissures. However, it is also quite possible that their use of a modified (Maudsley) 1020 electrode
array (Margerison et al., 1970), which gives electrodes a
better view of inferior brain convexities, effected downward displacement of the dipole. When combined with the
anticipated opposing effect of spherical forward modeling
in MTLE on dipole localization (upward displacement to
the frontal lobe), this might well result in partially compensated mislocalization to the retro-orbital region. For
ease of comparison, spike peaks were also modeled. If de
novo spikes were the product of discharge propagation,
then confining ESL to the spike peak may well exaggerate the influence of signal propagation on the final result.
Indeed, as the authors argue, the many discharge patterns
captured at the scalp relative to the deeper FO field were
the probable manifestation of a continuum of source behavior effectsfrom discharges seen only by the FO electrodes, to those seen more globally by virtue of volume
conduction, to those seen slightly later at the scalp by virtue
of signal propagation. The probing of such spatiotemporal
effects with a more sophisticated forwardinverse modeling set-up would have been very worthwhile, particularly
in light of the generally held view that deep mesial discharges are only seen at the scalp by virtue of propagation
alone (Ebersole, 2003a).
Finally, the authors did not show patient clinical data (radiology, pathology, and surgical outcome), which should
have been available, the presurgical work-up having been
performed between 1990 and 1998. Nevertheless, this work
demonstrates the value of using lower voltage spikes to
study interictal source behavior in MTLE and it challenges
the view that MTLE scalp spikes are the exclusive byproducts of cortico-cortical propagation from deeper mesial
structures.
Relatively few ictal studies addressing the correlation
between scalp and intracranial ESL have ever been published and, much like the previously described interictal
work, most studies have used presurgical TLE patient cohorts. The main ictal studies are limited to dipole modeling and arguably the most influential publication is 10

years old (Assaf & Ebersole, 1997). The authors examined the EEG recordings of 40 TLE patients who required
intracranial electrode implantation as part of the surgical work-up. All patients had an Engel class 1 outcome
with a mean follow-up of one year. Spatiotemporal dipole
modeling was carried out on scalp data that had been acquired with a 25-electrode set-up (standard 1020 set-up
plus three inferior temporal electrode pairs). Dipoles with
different orientations were preassigned to model different
sublobar divisions of the temporal cortical surface. The
investigators selected the dominant and/or leading dipole
model that explained the earliest recognizable, averaged
ictal rhythm seen at the scalp. Dipole models were then
matched with the ictal localization suggested by the intracranial recordings and expressed as positive predictive
values.
The results were impressive, with high positive predictive values found for the following source and seizure
onset match-ups: vertical tangential dipole (basal source)
and hippocampal onset (89%), horizontal tangential dipole
(temporal tip source) and entorhinal onset (83%), horizontal radial dipole (lateral source) and neocortical onset (80%). Multiple source components were modeled in
13 patients in whom an oblique dipole model (geometric
mean of above three source components) was thought to
explain seizure onset at the inferolateral temporal cortex.
As the authors indicated, the ictal ESL results are largely
in agreement with previous interictal ESL results in TLE.
This is perhaps not surprising given the relatively good correlation between interictal and ictal lateralization in TLE
(Blume et al., 2001). It should be noted that, as a retrospective study, scalp and intracranial EEG recordings were
not simultaneously acquired and the authors did not indicate that they were fully blinded to the intracranial data
when scalp ESL was performed. Also, many seizures were
captured (212 in total) but the statistical analysis was only
done on a patient-wise basis. Notwithstanding the Engel
class 1 outcome for all patients, a more rigorous analysis
of the consistency of dipole modeling for each seizure in
each patient would have been useful. Moreover, the investigators chose the dipole fit that best explained the early
ictal rhythm but did not describe the adequacy of the best fit
in more quantitative terms. For example, it is unclear how
well the dipole model explained the signal variance in each
case, or how well the leading dipole model dominated the
second dipole fit when their respective source components
overlapped spatiotemporally.
In a related publication (Assaf & Ebersole, 1999),
the authors showed how their dipole modeling approach
might be used to anticipate surgical success following either standard or modified anteromesial temporal lobectomy (AMTL). After a minimum follow-up period of two
years, they found that postsurgical seizure freedom was
more likely to occur in patients whose dipole modeling
suggested a dominant or leading basal source, but was
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

12
C. Plummer et al.
most likely to occur in patients whose dipole modeling
did not implicate a lateral source. The latter association
was explained on the basis that neocortical foci are unlikely to be resected adequately by AMTL, even in the
modified approach when the lateral surgical margin is
extended.
Of the few ictal ESL studies that have compared simultaneously acquired scalp and intracranial EEG in focal
epilepsy, it is difficult to look past the methodical study by
Merlet and Gotman in which the accuracy of spatiotemporal dipole modeling from scalp EEG (2840 channel setup) was judged in relation to the localization suggested
by depth and/or epidural electrode recordings (Merlet &
Gotman, 2001). A consecutive series of 15 presurgical patients with various forms of refractory partial epilepsy were
enrolled in the study. Patients were excluded from ESL if
they did not have at least two reproducible seizure patterns
recorded at the scalp, one following the other. By specifying this condition, the investigators are the first to have systematically examined the spatial resolution of dipole localization as the seizure pattern evolves at the scalp electrodes.
Of the nine patients (seven lesion-negative) who met this
condition, six patients (five lesion-negative) had averaged
ictal patterns that could produce a sufficiently stable and
interpretable ESL result. As noted by other investigators,
the earliest ictal rhythm seen at the scalp was always associated with intracranial discharges occupying large areas
of cortex. In a new finding though, the three patients with
unstable ESL results had, by the time an ictal rhythm was
seen at the scalp, an intracranial ictal rhythm that was bilateral with maximal amplitude at the mesial temporal region.
Further, of the six patients who returned a stable ESL result
at some point during the seizure, only half had scalp EEG
changes that were concomitant with seizure onset as suggested by the intracranial recording. The dominant source
in each case implicated a neocortical temporal focus with
distances from the main dipole to the nearest (maximal
amplitude) intracranial electrode as follows: 5.5 mm, 8.3
mm, and 20.8 mm. Two of these patients, and two others
(four of six), had dominant dipoles that coincided with the
locations of intracranially recorded voltage maxima during
the earlier ictal pattern, but this was only the case for two
of the six patients when the second (later) ictal pattern was
modeled.
While this study represents one of the most carefully
conducted analyses of ESL to date, the interpretation of the
findings was perhaps slightly misdirected. Much emphasis
was given to the physical separation of the dominant dipole
to dominant electrode in trying to validate the scalpderived ESL result. The authors did point out that their
quoted distances should not be regarded as error margins,
but rather as measures of concordance. However, measures
of spatial inconsistency for ESL results ranged from
15.3 mm to 38.4 mm. These distances are modest when
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

one considers the following factors: that intracranial EEG


localization carries an inherent error due to the problem
of under-sampling the cortical garland; that dipoles model
cortical surfaces in centimeter dimensions; that dipole orientation carries far more weight than dipole location in
modeling source origin and/or propagation; and that the
amplitude of the ictal discharge, on which the measurements were based, is highly dependent on the orientation of
the actual source as it faces the recording electrodes. To
illustrate, the authors concluded that mesial onset seizures
are prone to mislocalization by scalp ESL based on their
concordance measurements from the amygdala (maximal
intracranial signal) to the anterior temporal region (dominant dipole location) in two patients (20.8 mm and 38.4
mm). However, such distances are not unexpected in light
of the recognized patterns of ictal and interictal propagation in MTLE. More useful would have been a clear description of dipole orientation for the spatiotemporal models applied to each ictal pattern for these two patients,
but ideally for all patients. Because scalp-derived ESL localization is more likely to reflect discharge propagation,
versus discharge onset, assessing ESL spatial accuracy by
location of the intracranial EEG maxima is problematic.
Once again, it is the orientation of the dipoleand the extended area of cortex to which that dipole projectsthat
yields a truer indication of ESL validity in modeling ictal
or interictal patterns of onset and propagation.
Lastly, the investigators used a relatively low high-pass
filter setting (0.3 Hz) and applied start markers for their
ictal modeling at a latency well before any deflection was
noted in the scalp trace (based on a figure supplied). It is
therefore possible that early ictal signal quality may have
been compromised for ESL, leading to noisier, less-stable
dipole solutions.
It should be emphasized here that the use of intracranial EEG recordings as a gold standard validation tool
for ESL is problematic. However meticulously conducted,
scalp-intracranial EEG studies, such as those discussed
above, are necessarily limited by the extent to which the
cortex is sampled at depth. For instance, FO electrodes
only sample the entorhinal cortex and orthogonal depth arrays, as used in the last study (Merlet & Gotman, 2001),
have a limited view of the temporal neocortex. Separating
source origin from propagation will, in such circumstances,
involve a measure of speculation on the part of the investigator. The above ESL findings await replication, ideally
with the use of electrode arrays that sample larger areas of
cortex; as with sampling that sufficiently includes mesial,
inferior, anterior tip, and lateral temporal lobe cortical surfaces in the case of TLE interictal and ictal ESL. In fact,
without such studies, the capacity for ESL to help characterize the interplay between the lesion, irritative zone, the
seizure onset zone, and the epileptogenic zone will remain
untested.

13
EEG Source Localization in Focal Epilepsy

OTHER N ONINVASIVE S OURCE


L OCALIZATION M ETHODS
If publication output is anything to go by, MEG and
EEG-fMRI have attracted greater research interest than that
enjoyed by ESL in recent times. Signal acquisition techniques have improved for both methods, which promise
better spatial resolution than that offered by currentgeneration ESL. For MEG, modern high-density sensor arrays that encompass the whole head, and not just part of
it, have allowed recordings to be done in a single step,
rather than in a cumbersome, piece-wise manner. For EEGfMRI, novel ways of recording EEG in the hostile MRI
environment have given researchers the chance to better
understand how cortical and subcortical hemodynamic response patterns are coupled to scalp recorded spike and
seizure patterns. For ESL then, there are two immediate
questions.
Has ESL been marginalized by MEG?
It should be emphasized that the principles of signal
detection and source localization for EEG apply just as
equally to MEG. That is, signal detection in MEG depends on the recruitment of a sufficient population of discharging cortical neurons that are synchronized in time
and aligned in space. And for source localization in MEG,
both the forward and inverse problems need to be resolved
with suitable models to achieve a tenable solution. This
factthat MEG-based source localization is answerable
to the same kind of fundamental mathematical problems
that underpin ESLis easily overwhelmed by the glare
of the technology on offer. MEG technology is seductive but expensive. Nunez puts it another way. Enthusiasm for MEG (relative to EEG) has been boosted by both
genuine scientific considerations and poorly justified commercial pressures (Nunez & Srinivasan, 2006b). Even if
MEG becomes more portable and affordable at some future date, ESL is likely to maintain an important stake in
noninvasive source modeling. This is because, apart from
the obvious practical advantages currently held by EEG
over MEG (pediatric and ictal studies, long-term monitoring are largely prohibited by head movement artifacts),
EEG and MEG see spike and seizure discharges quite
differently.
The practice of pitting MEG against EEG in the source
imaging literature has perhaps been overdone, and there
is now an evolving consensus that the combined use of
these techniques (in the rare situation when both are accessible) optimizes source localization accuracy (Fuchs et
al., 1998; Barkley & Baumgartner, 2003). While comparisons between the two methods will continue to be made,
it should be noted that to date no study has been published which compares source modeling accuracy for simultaneously acquired MEG and EEG recordings against
simultaneously acquired intracranial data (as the surrogate

gold standard) in a prospective, blinded manner for focal


epilepsy.
A common misconception in this regard is that MEG is
more accurate than EEG in defining source activity owing to its superior spatial resolution and its relative immunity to field distortion by volume conductor effects (Nunez
& Srinivasan, 2006b). While these latter qualities are true
enough, MEG only picks up part of the cortical activity
generated by the source(s). To coin an analogy, one might
imagine that if a patch of discharging cortex is likened to
a 10-cm2 piece of undulating cheese, EEGs field of view
will be blurred and distorted by the overlying cellophane
wrapper, while MEGs view will be clearer, but restricted,
as if the cheese had been made Swiss. This is because
MEG is blind to the radial vector component of the electric field. Thus, the magnetic field is less complicated
by variably admixed radial and tangential vectors and it is
less distorted by the skullscalp interface. However, MEGbased modeling has an inherent bias for superficial sources
because the magnetic field decays very rapidly from scalp
surface (Hamalainen et al., 1993).
It is interesting to note that MEG source imaging (MSI)
researchers have almost exclusively applied single fixed
dipoles to model spikes in focal epilepsy. While the use of
such a simple inverse algorithm in MSI might better suit the
modeling of the cleaner signal topography of magnetic
versus electrical fields, it is a mistake to think that MEG is
immune from the same properties of signal propagation as
EEG. Gloor puts it bluntly. Modeling the magnetic field
based on the assumption that the field can be represented
by a single fixed dipole is fraught with difficulties similar
to those inherent in the modeling of electrical fields based
on this assumption (Gloor, 1985).
Therefore, the interpretation of MEG source localization
accuracy on the millimeter scale, as is the usual case in the
MSI literature, should be eyed with at least a degree of caution. Indeed, the area of cortex required for contemporary
MEG arrays to detect an interpretable field is still in the order of 3 cm2 at best (Oishi et al., 2002). A recent paper by
Fischer and colleagues has looked to redress this issue with
the calculation of ellipsoid volumes based on dipole cluster
variability for a population of spikes in a presurgical group
(Fischer et al., 2005).
It is repeatedly emphasized in the MSI literature that,
although blind to radial source components (from gyral
crests), most of the cortex is seen by MEG arrays because,
as observed by Brodmann nearly 100 years ago, the cumulative gyral surface only accounts for a third of the total
surface area of the human brain (Brodmann, 1909). But
the implication here is that anatomy and physiology are
measured with the same ruler. As Wong reminds us, cortical lamination, pyramidal arborization, cortical vascularization, and cortico-cortical connections are richer at gyral
crowns than at fissural walls (Wong, 1998). Gyri also account for much of the cortical homunculus in man (Welker,
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

14
C. Plummer et al.

Figure 4.
Demonstration of the relationships between cortical anatomy, surface electrical field, and dipole model for a right
frontotemporal spike in a patient with focal epilepsy. T 1 -weighted MRI coronal (A) and sagittal (B) views; and surface
rendered cortical images, oblique (C) and from above (D), are shown. Note that dipoles (rotating over 40 msec
epoch) and the associated confidence ellipsoid volumes do not conform to either the position or orientation of a
particular sulcus, gyrus, or even lobe (A, B). Rather, dipoles model the summated electrical field recorded at the
scalp electrodes (surface negative over right hemisphere and surface positive over left hemisphere in this case, C),
and they assume orientations and positions that attempt to explain the polarity characteristics of this surface field
(D). Therefore, large areas of cortex, comprising multiple sulci and gyri, are represented electrically in the modeling
of interictal or ictal events. It is the net configuration (orientation, position, and strength) of micro dipolar fields
(variably seen by the surface electrodes as they project orthogonally from numerous, individual gyral and sulcal
surfaces) that is ultimately modeled by the dipole solution. Abbreviations: MRI (Magnetic Resonance Image).
C ILAE
Epilepsia !
1990; Wong, 1998). Much of the immediate fissural region
is dedicated to propagation of interictal/ictal discharges,
a phenomenon that is generally ill suited to single fixed
dipole modeling strategies as discussed earlier. This is especially so if modeling is restricted to the spike peak, a
practice commonly adhered to in MEG epilepsy studies.
The redundancy of the anatomical two-thirds argument
is made clearer when it is recalled that dipoles are not
anatomically based constructs, but representations of the
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

summated electrical field generated by areas of cortex large


enough to include both gyral and fissural surfaces (Fig. 4).
Despite these caveats, support for MSI as a legitimate
source localization technique has grown. Many clinical
studies, most of which are understandably interictal, have
found favorable correlations between MSI location and
the localization suggested by either intracranial recordings or postsurgical seizure recurrence ratesfor keynote
studies see (Stefan et al., 2003; Pataraia et al., 2004) and

15
EEG Source Localization in Focal Epilepsy
for reviews see (Ebersole, 1997b; Barkley, 2004). However, as pointed out in a more recent review (Makela et
al., 2006), no large prospective trial incorporating a randomized control group (with sham lesion margins) has
been carried out. Most trials have been performed retrospectively on small patient groups when investigators have
not been blinded to the patients corresponding MRI data.
Similarly, levels of intra- and interoperator concordance
for MSI localization results in these studies have not been
clearly assessed.
Can EEG-fMRI optimize the spatial resolution
of ESL?
The ability to correlate the interictal waveform of EEG
with the blood oxygen level dependent (BOLD) response
of fMRI is a relatively recent development in functional
imaging in epilepsy (Seeck et al., 1998; Goldman et al.,
2000; Krakow et al., 2000). Several authors have proposed
that simultaneously acquired fMRI should enhance the spatial resolution of ESL (Liu et al., 1998; Krakow et al., 1999;
Phillips et al., 2002). The idea is that spike-correlated fMRI
data can be used to constrain the ESL solution to one or
other regions of interest. Such a constraint might be expected to hold more physiological credence than traditional
dipole or distributed ESL constraints that are founded on
relatively pure mathematics.
However, the EEG-fMRI marriage has not been so cosy.
Accumulating evidence suggests that the spatiotemporal
relationship between the two is far from straightforward.
To begin with, the temporal resolution of fMRI, which is
dependent on the properties of blood flow and blood deoxygenation, lags EEG by a factor of 1,000. This means
that the electrical field and the hemodynamic response generated by an epileptiform discharge must be coupled by
an archetypal time-constant, the so-called hemodynamic
response function (HRF). Aside from the variability in
the HRF that can occur between individuals (Aguirre et
al., 1998) and between spike populations (Bagshaw et al.,
2004), the validity of a one-size-fits-all HRF must be questioned when the pathology presumed to be responsible for
the patients epilepsy interferes with the regional integrity
of the normal bloodbrain barrier (viz. gliosis, oedema,
vascular malformations). And, despite the promising spatial resolution of fMRI, (which, it must be remembered, is
heavily influenced by operator-dependent thresholding of
the signal), the relative temporal blurring of the BOLD response makes it virtually impossible to tease out discharge
onset from discharge propagation in contemporary EEGfMRI recordings.
The disjointed spatiotemporal relationship between EEG
and fMRI is underscored by the fact that the fMRI BOLD
signal is a physiological response tied to brain structure and
one that is not dependent on the synchronized and aligned
neuronal activity on which both EEG and MEG are based.
Because scalp EEG is used to signpost interictal and ic-

tal events for interpretation of the BOLD, one is left with


analyzing only that fraction of the BOLD activity that has
been lassoed, however loosely via the HRF, to electrical
events visible at the scalp. Electrical events unseen at the
scalpand estimates of simultaneously detectable surface:
depth spikes have been put as low as 1:2000(Alarcon et
al., 1994) are, by default, forfeited when it comes to the
fMRI BOLD analysis. The converse approach, that is, using ESL to understand the significance of BOLD activity in focal epilepsy, seems more logical. A recent EEGfMRI study in benign partial epilepsy with centrotemporal
spikes suggests that multiple spatiotemporal dipole modeling can help unravel the temporal sequence of BOLD activation associated with interictal discharges (Boor et al.,
2007). In this study, centrotemporal spikes were characterized by an initial dipole corresponding to BOLD activation
at the central sulcus; and a second, later dipole corresponding to BOLD activation at the sylvian fissure. Based on the
ESL sequence, the pattern of BOLD activation was thought
to capture the spatial onset and propagation of the corresponding electrical event.
Still adding to the complexity of the EEG-fMRI relationship is the observation that both neuronal excitation
and inhibition can lead to an increase in the BOLD signal (Gotman et al., 2006). This is because the BOLD
signal is thought to reflect neuronal energy consumption in
a broad sense. Less intuitive then, is the significance of a
decrease in the BOLD signal, or deactivation. It has been
proposed (Kobayashi et al., 2006) that deactivation might
arise from a generalized state of functional de-afferentation
(akin to a state of suspended animation); or, from a relative increase in inhibition occurring at the level of the soma,
a process that is less energy consuming than inhibition occurring at the level of the postsynaptic membrane.
In short then, given our incomplete grasp of the fundamental determinants of the BOLD response in EEG-fMRI
signal acquisition, and given the relatively inharmonious
relationship between EEG and fMRI signal dynamics, the
use of fMRI as a default ESL constraint cannot be recommended at this stage. Much like MSI, more can be gained
by respecting the independence of the two methods for
source localization in focal epilepsy. Indeed, the converse,
the use of ESL with EEG-fMRI to rationalize the temporal
dynamics of fMRI, may be a more fruitful approach.

C ONCLUSION
This review has examined the current state of EEG
source localization (ESL) as it applies to focal epilepsy.
We have outlined the important principles that guide the
methodological approach to contemporary ESL and we
have highlighted some of the common misconceptions related to dipole and distributed modeling.
Based on the progress made in clinically directed research in the last 510 years, we believe that ESL will
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

16
C. Plummer et al.
earn a place in the routine work-up of patients with focal epilepsy in the foreseeable future. To get there, we anticipate that the following conditions need to be met: (a)
blinded, prospective validation studies conducted on larger
clinical groups; (b) the more routine use of realistic forward models; (c) a greater awareness of the importance of
source orientation in defining inverse modeling solutions,
with elimination of the tendency to use source location as
the cardinal determinant of ESL accuracy; (d) the avoidance of restricting source modeling to spike peaks such that
earlier spike components are routinely included in the ESL
analysis; (e) the avoidance of using zero-phase shift filtering on EEG signals; (f) the inclusion of error measures,
such as confidence ellipsoid volumes, into ESL solutions to
improve the statistical robustness of clinically based findings; and (g) when permitted by the SNR, clarification of
the likely clinical impact of choosing single versus averaged spike epochs for ESL.
We have argued that ESL has by no means been swept
aside by the more advanced functional imaging modalities of MEG and EEG-fMRI, principally because the three
modalities emphasize quite different bio-electromagnetic
phenomena. As with all investigative modalities used in
the clinical work-up of patients with focal epilepsy, including intracranial EEG localization, each is beset by its
own problems in trying to define the epileptogenic zone.
This is perhaps a better way to judge the relative merits
of source localization strategiesby understanding the respective limitations of each method from the start.
With this approach, loose assumptions and dogmatic
conclusions are less likely to be made. In fact, the leading source localization approach, or the one that should
be most valued, has not changed much in nearly a century despite the relatively recent boom in anatomical and
functional imaging in epilepsythe scalp recorded EEG in
the context of a rigorous patient history. The multimodal
imaging technology on offer today for spike and seizure
localization has limited value if its results are interpreted
without due respect for the patients electroclinical seizure
manifestations. In this sense, ESL is better regarded as a
much underutilized three-dimensional extension of traditional two-dimensional EEG analysis. As a technology that
is now quite accessible, affordable, noninvasive, and one
that is founded on the well-established electrophysiological principles of EEG, ESL continues to hold promise as
a potential clinical tool that offers to teach us much about
the recruitment and propagation of interictal and ictal discharges in focal epilepsy.

ACKNOWLEDGEMENTS
The authors would like to thank Dr. Michael Wagner and Dr.
Manfred Fuchs for their patience and support in explaining (and
reexplaining!) the often-vexed mathematics that goes hand-inhand with EEG and MEG source localization; and Mr. Simon
Vogrin, Mr. Lucas Litewka, and Dr Kevin Morris for their enEpilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

couragement and debate on the topic. The first author would like
to personally thank Dr. John Ebersole for his many lively presentations on dipole modeling. Perhaps more than anyone else in the
area, he has strived to keep the discipline grounded in its clinical
roots while always endeavoring to keep the mathematicians and
physicists honest!
The first author holds an Australian National Health and Medical Research Council Postgraduate Medical Research Scholarship. We confirm that we have read the Journals position on issues involved in ethical publication and we affirm that this report
is consistent with those guidelines. There are no disclosures from
any of the authors.

R EFERENCES
Aguirre GK, Sarah E, Depositor M. (1998) The variability of human,
BOLD homodynamic responses. Neuroimage 8:360369.
Alarcon G, Guy CN, Binnie CD, Walker SR, Elwes RDC, Polkey
CE. (1994) Intracerebral propagation of interictal activity in partial
epilepsyimplications for source localization. J. Neurol. Neurosurg.
Psychiatry 57:435449.
Assaf BA, Ebersole JS. (1997) Continuous source imaging of scalp ictal
rhythms in temporal lobe epilepsy. Epilepsia 38:11141123.
Assaf BA, Ebersole JS. (1999) Visual and quantitative ictal EEG predictors of outcome after temporal lobectomy. Epilepsia 40:5261.
Bagshaw AP, Aghakhani Y, Benar CG, Kobayashi E, Hawco C, Dubeau
F, Pike GB, Gotman J. (2004) EEG-fMRI of focal epileptic spikes:
analysis with multiple haemodynamic functions and comparison with
gadolinium-enhanced MR angiograms. Hum Brain Map 22:179192.
Barkley GL. (2004) Controversies in neurophysiology. MEG is superior
to EEG in localization of interictal epileptiform activity: pro. Clin
Neurophysiol 115:10011009.
Barkley GL, Baumgartner C. (2003) MEG and EEG in epilepsy. J Clin
Neurophysiol 20:163178.
Benar C, Gotman J. (2001) Non-uniform spatial sampling in EEG source
analysis. Engineering in Medicine and Biology Society, 2001. Proceedings of the 23rd Annual International Conference of the IEEE
1:903905.
Binnie CD, Dekker E, Smit A, Van Der Linden G. (1982) Practical considerations in the positioning of EEG electrodes. Electroencephalogr
Clin Neurophysiol 53:453458.
Blume WT, Holloway GM, Wiebe S. (2001) Temporal epileptogenesis:
localizing value of scalp and subdural interictal and ictal EEG data.
Epilepsia 42:508514.
Boon P, DHave M, Vanrumste B, Van Hoey G, Vonck K, Van Walleghem
P, Caemaert L, Achten E, De Reuck J. (2002) Ictal source localization
in presurgical patients with refractory epilepsy. J Clin Neurophysiol
19:461468.
Boor R, Jacobs J, Hinzmann A, Bauermann T, Scherg M, Boor S, Vucurevic G, Pfleiderer C, Kutschke G, Stoeter P. (2007) Combined
spike-related functional MRI and multiple source analysis in the noninvasive spike localization of benign rolandic epilepsy. Clin Neurophysiol 118:901909.
Braga NI, Manzano GM, Nobrega JA. (2002) A comparison between
averaged spikes and individual visually-analyzed spikes in rolandic
epileptiform discharges. Arq Neuropsiquiatr 60:699701.
Braitenberg V. (1978) Cortical architechtonics: general and areal. In Brazier MAB, Petsche H (Eds) Architechtonics of the cerebral cortex,
Raven Press, New York, pp. 443465.
Brazier MAB. (1949) A study of the electrical fields of the surface of the
head. Electroencephal Clin Neurophysiol 2:3852.
Brodmann K. (1909) Vergleichende Lokalisationslehre der Grosshirnrinde. Barth, Leipzig.
Buzsaki G, Traub RD, Pedley TA. (2003) The cellular basis of EEG activity. In Ebersole JS, Pedley TA (Eds) Current practice of clinical
electroencephalography, Lippincott Williams and Wilkins, Philadelphia, pp. 111.
Chabardes S, Kahane P, Minotti L, Tassi L, Grand S, Hoffmann D, Benabid AL. (2005) The temporopolar cortex plays a pivotal role in temporal lobe seizures. Brain 128:18181831.

17
EEG Source Localization in Focal Epilepsy
Chitoku S, Otsubo H, Ichimura T, Saigusa T, Ochi A, Shirasawa A,
Kamijo K, Yamazaki T, Pang E, Rutka JT, Weiss SK, Snead OC.
(2003) Characteristics of dipoles in clustered individual spikes and
averaged spikes. Brain Develop 25:1421.
Crouzeix A, Yvert B, Bertrand O, Pernier J. (1999) An evaluation of
dipole reconstruction accuracy with spherical and realistic head models in MEG. Clin Neurophysiol 110:217688.
Cuffin BN. (1996) EEG localization accuracy improvements using realistically shaped head models. IEEE Trans Biomed Eng 43:299303.
Darvas F, Pantazis D, Kucukaltun-Yildirim E, Leahy RM. (2004) Mapping human brain function with MEG and EEG: methods and validation. Neuroimage 23:S289S299.
Ebersole J, Wade P. (1990) Spike voltage topography and equivalent
dipole localization in complex partial epilepsy. Brain Topogr 3:21
34.
Ebersole JS. (1994) Noninvasive localization of the epileptogenic focus
by EEG dipole modeling. Acta Neurol Scand 89:2028.
Ebersole JS. (1997a) Defining epileptogenic foci: past, present, future. J
Clin Neurophysiol 14:470483.
Ebersole JS. (1997b) Magnetoencephalography/magnetic source imaging
in the assessment of patients with epilepsy. Epilepsia 38:S1S5.
Ebersole JS. (2000) Noninvasive localization of epileptogenic foci by
EEG source modeling. Epilepsia 41(Suppl 3):S24S33.
Ebersole JS. (2003a) EEG voltage topography and dipole source modeling of epileptiform potentials. In Ebersole JS, Pedley TA (Eds) Current practice of clinical electroencephalography, Lippincott Williams
and Wilkins, Philadelphia, pp. 732752.
Ebersole JS. (2003b) Cortical generators and EEG voltage fields. In
Ebersole JS, Pedley TA (Eds) Current practice of clinical electroencephalography, Lippincott Williams and Wilkins, Philadelphia, pp.
1231.
Ebersole JS, Hawes-Ebersole S. (2007) Clinical application of dipole
models in the localization of epileptiform activity. J Clin Neurophysiol 24:120129.
Fischer MJM, Scheler G, Stefan H. (2005) Utilization of magnetoencephalography results to obtain favourable outcomes in epilepsy
surgery. Brain 128:153157.
Fuchs M, Wagner M, Wischmann HA, Kohler T, Theissen A, Drenckhahn
R, Buchner H. (1998) Improving source reconstructions by combining
bioelectric and biomagnetic data. Electroencephal Clin Neurophysiol
107:93111.
Fuchs M, Wagner M, Kohler T, Wischmann HA. (1999) Linear and nonlinear current density reconstructions. J Clin Neurophysiol 16:267
295.
Fuchs M, Kastner J, Wagner M, Hawes S, Ebersole JS. (2002) A standardized boundary element method volume conductor model. Clin Neurophysiol 113:702712.
Fuchs M, Ford MR, Sands S, Lew HL. (2004a) Overview of dipole source
localization. Phys Med Rehabil Clin N Am 15:251262.
Fuchs M, Wagner M, Kastner J. (2004b) Confidence limits of dipole
source reconstruction results. Clin Neurophysiol 115:14421451.
Fuchs M, Wagner M, Kastner J. (2007) Development of volume conductor and source models to localize epileptic foci. J Clin Neurophysiol
24:101119.
Geddes LA, Baker LE. (1967) The specific resistance of biological
material- a compendium of data for the biomedical engineer and physiologist. Med Biol Eng 5:271293.
Gevins A. (1993) High resolution EEG. Brain Topogr 5:321325.
Gloor P. (1985) Neuronal generators and the problem of localization in
electroencephalography: application of volume conductor theory to
electroencephalography. J Clin Neurophysiol 2:327354.
Goldman RI, Stern JM, Engel J, Jr., Cohen MS. (2000) Acquiring simultaneous EEG and functional MRI. Clin Neurophysiol 111:1974
1980.
Gotman J. (2003) Noninvasive methods for evaluating the localization and
propagation of epileptic activity. Epilepsia 44:2129.
Gotman J, Kobayashi E, Bagshaw AP, Benar CG, Dubeau F. (2006) Combining EEG and fMRI: a multimodal tool for epilepsy research. J
Magn Reson Imaging 23:906920.
Grave de Peralta Menendez R, Gonzalez Andino S, Lantz G, Michel
CM, Landis T. (2001) Noninvasive localization of electromagnetic
epileptic activity. I. Method descriptions and simulations. Brain Topogr 14:131137.

Ha KS, Youn T, Kong SW, Park HJ, Ha TH, Kim MS, Kwon JS. (2003)
Optimized individual mismatch negativity source localization using a
realistic head model and the Talairach coordinate system. Brain Topogr 15:233238.
Hamalainen MS, Hari R, Ilmoniemi RJ, Knuutila J, Lounasmaa OV.
(1993) Magnetoencephalography: theory, instrumentation, and applications to noninvasive studies of the working human brain. Rev Modern Phys 65:413497.
Hamalainen MS, Ilmoniemi RJ. (1994) Interpreting magnetic fields of the
brain: minimum norm estimates. Med Biol Eng Comput 32:3542.

Helmholtz H. (1853) Uber


die Methoden, kleinste Zeitteile zu messen,
und ihre Anwendung fur physiologische Zwecke. Original work translated in. Philosophical Magazine 6:313325.
Jasper H (1969) Mechanisms of propagation: extracellular studies. In
Jasper H, Ward A, Pope A (Eds) Basic mechanisms of the epilepsies,
Little, Brown & Co., Boston, pp. 421438.
Jayakar P, Duchowny M, Resnick TJ, Alvarez LA. (1991) Localization of
seizure focipitfalls and caveats. J Clin Neurophysiol 8:414431.
Knott JR (1985) Further thoughts on polarity, montages, and localization.
J Clin Neurophysiol 2:6375.
Kobayashi E, Bagshaw AP, Grova C, Dubeau F, Gotman J. (2006) Negative BOLD responses to epileptic spikes. Hum Brain Mapp 27:488
497.
Krakow K, Woermann FG, Symms MR, Allen PJ, Lemieux L, Barker
GJ, Duncan JS, Fish DR. (1999) EEG-triggered functional MRI of
interictal epileptiform activity in patients with partial seizures. Brain
122:16791688.
Krakow K, Allen PJ, Lemieux L, Symms MR, Fish DR. (2000) Methodology: EEG-correlated fMRI. Adv Neurol 83:187201.
Krauss GL, Webber WRS. (2005) Digital EEG. In Niedermeyer E, Lopes
Da Silva F (Eds) Electroencephalography: basic principles, clinical applications, and related fields. Lippincott Williams & Wilkins,
Philadelphia, pp. 797814.
Lantz G, Grave de Peralta Menendez R, Gonzalez Andino S, Michel CM
(2001) Noninvasive localization of electromagnetic epileptic activity.
II. Demonstration of sublobar accuracy in patients with simultaneous
surface and depth recordings. Brain Topogr 14:139147.
Lantz G, de Peralta RG, Spinelli L, Seeck M, Michel CM. (2003a) Epileptic source localization with high density EEG: how many electrodes
are needed? Clin Neurophysiol 114:6369.
Lantz G, Spinelli L, Seeck M, Menendez RGD, Sottas CC, Michel CM
(2003b) Propagation of interictal epileptiform activity can lead to erroneous source localizations: a 128-channel EEG mapping study. J
Clin Neurophysiol 20:311319.
Lehmann D (1987) Principles of spatial analysis. In Gevins AS, Remond
A (Eds) Methods of analysis of brain electrical and magnetic signals.
Elsevier, Amsterdam, pp. 309354.
Liu AK, Belliveau JW, Dale AM. (1998) Spatiotemporal imaging of
human brain activity using functional MRI constrained magnetoencephalography data: Monte Carlo simulations. Proc Natl Acad Sci
U S A 95:89458950.
Makela LP, Forss N, Jaaskelainen J, Kirveskari E, Korvenoja A, Paetau
R. (2006) Magnetoencephalography in neurosurgery. Neurosurgery
59:493510.
Margerison JH, Binnie CD, McCaul IR. (1970) Electroencephalographic
signs employed in the location of ruptured intracranial arterial
aneurysms. Electroencephalogr Clin Neurophysiol 28:296306.
Merlet I, Gotman J. (1999) Reliability of dipole models of epileptic
spikes. Clin Neurophysiol 110:10131028.
Merlet I, Gotman J. (2001) Dipole modeling of scalp electroencephalogram epileptic discharges: correlation with intracerebral fields. Clin
Neurophysiol 112:414430.
Michel CM, Murray MM, Lantz G, Gonzalez S, Spinelli L, de Peralta RG.
(2004a) EEG source imaging. Clin Neurophysiol 115:21952222.
Michel CM, Lantz G, Spinelli L, De Peralta RG, Landis T, Seeck M.
(2004b) 128-channel EEG source imaging in epilepsy: clinical yield
and localization precision. J Clin Neurophysiol 21:7183.
Mori S, van Zijl PC. (2002) Fiber tracking: principles and strategiesa
technical review. NMR Biomed 15:468480.
Nayak D, Valentin A, Alarcon G, Seoane JJG, Brunnhuber F, Juler J,
Polkey CE, Binnie CD (2004) Characteristics of scalp electrical fields
associated with deep medial temporal epileptiform discharges. Clin
Neurophysiol 115:14231435.
Epilepsia, **(*):118, 2007
doi: 10.1111/j.1528-1167.2007.01381.x

18
C. Plummer et al.
Niedermeyer E. (2005) The EEG signal: polarity and field determination.
In Niedermeyer E, Lopes Da Silva F (Eds) Electroencephalography:
basic principles, clinical applications, and related fields. Lippincott
Williams & Wilkins, Philadelphia, pp. 125130.
Nunez P, Srinivasan R. (2006a) Electric fields and currents in biological
tissue. In Nunez P, Srinivasan R (Eds.) Electric fields of the brain.
The neurophysics of EEG, Oxford University Press, New York, pp.
147202.
Nunez P, Srinivasan R. (2006b) Fallacies in EEG. In Nunez P, Srinivasan
R (Eds) Electric fields of the brain. The neurophysics of EEG, Oxford
University Press, New York, pp. 5698.
Oishi M, Otsubo H, Kameyama S, Morota N, Masuda H, Kitayama M, Tanaka R. (2002) Epileptic spikes: magnetoencephalography versus simultaneous electrocorticography. Epilepsia 43:1390
1395.
Pascual-Marqui RD, Michel CM, Lehmann D. (1994) Low resolution
electromagnetic tomography: a new method for localizing electrical
activity in the brain. Int J Psychophysiol 18:4965.
Pataraia E, Simos PG, Castillo EM, Billingsley RL, Sarkari S, Wheless JW, Maggio V, Maggio W, Baumgartner JE, Swank PR, Breier
JI, Papanicolaou AC. (2004) Does magnetoencephalography add to
scalp video-EEG as a diagnostic tool in epilepsy surgery? Neurology
62:943948.
Phillips C, Rugg MD, Friston KJ. (2002) Anatomically informed basis functions for EEG source localization: combining functional and
anatomical constraints. Neuroimage 16:678695.
Plummer C, Litewka L, Farish S, Harvey AS, Cook MJ. (2007)
Clinical utility of current-generation dipole modelling of scalp
EEG. Neuropsysiology E-pub ahead of print doi:10.1016j.clinph.2007.08.016<http://dx.doi.org/10.1016/j.clinph.2007.08.016>.
Rosenow F, Luders H. (2001) Presurgical evaluation of epilepsy. Brain
124:16831700.
Roth BJ, Ko D, vonAlbertiniCarletti IR, Scaffidi D, Sato S. (1997) Dipole
localization in patients with epilepsy using the realistically shaped
head model. Electroenceph Clin Neurophysiol 102:159166.
Scherg M, Bast T, Berg P. (1999) Multiple source analysis of interictal
spikes: goals, requirements, and clinical value. J Clin Neurophysiol
16:214224.

Epilepsia, **(*):118, 2007


doi: 10.1111/j.1528-1167.2007.01381.x

Seeck M, Lazeyras F, Michel CM, Blanke O, Gericke CA, Ives J,


Delavelle J, Golay X, Haenggeli CA, de Tribolet N, Landis T.
(1998) Non-invasive epileptic focus localization using EEG-triggered
functional MRI and electromagnetic tomography. Electroencephalogr
Clin Neurophysiol 106:508512.
Sperli F, Spinelli L, Seeck M, Kurian M, Michel CM, Lantz G. (2006)
EEG source imaging in pediatric epilepsy surgery: a new perspective
in presurgical workup. Epilepsia 47:981990.
Srinivasan R, Nunez PL, Tucker DM, Silberstein RB, Cadusch PJ. (1996)
Spatial sampling and filtering of EEG with spline Laplacians to estimate cortical potentials. Brain Topogr 8:355366.
Stefan H, Hummel C, Scheler G, Genow A, Druschky K, Tilz C,
Kaltenhauser M, Hopfengartner R, Buchfelder M, Romstock J. (2003)
Magnetic brain source imaging of focal epileptic activity: a synopsis
of 455 cases. Brain 126:23962405.
Tao JX, Ray A, Hawes-Ebersole S, Ebersole JS. (2005) Intracranial
EEG Substrates of Scalp EEG Interictal Spikes. Epilepsia 46:669
676.
Wagner M, Kohler T, Fuchs M, Kastner J. (2001) An extended source
model for current density reconstructions. In Nenonen J, Ilmoniemi
RJ, Katila T (Eds.) Biomag 2000: Proceedings of the 12th International Conference on Biomagnetism, Helsinki University of Technology, Finland, pp. 749752.
Welker WI. (1990) Why does cerebral cortex fissure and fold? Cereb Cortex 8:3136.
Wilson FN, Bayley RH. (1950) The electric field of an eccentric dipole
in a homogeneous spherical conducting medium. Circulation 1:84
92.
Wong PK. (1998) Potential fields, EEG maps, and cortical spike generators. Electroenceph Clin Neurophysiol 106:138141.
Zumsteg D, Friedman A, Wennberg RA, Wieser HG. (2005) Source localization of mesial temporal interictal epileptiform discharges: correlation with intracranial foramen ovale electrode recordings. Clin Neurophysiol 116:28102818.
Zumsteg D, Friedman A, Wieser HG, Wennberg RA. (2006) Propagation of interictal discharges in temporal lobe epilepsy: correlation
of spatiotemporal mapping with intracranial foramen ovale electrode
recordings. Clin Neurophysiol 117:26152626.

También podría gustarte