Content uploaded by Jan C De Munck
Author content
All content in this area was uploaded by Jan C De Munck on Dec 28, 2018
Content may be subject to copyright.
This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Technical Note
A semi-automatic method to determine electrode positions and labels from gel
artifacts in EEG/fMRI-studies
Jan C. de Munck
a,
⁎, Petra J. van Houdt
a,b
, Ruud M. Verdaasdonk
a
, Pauly P.W. Ossenblok
b
a
VU University Medical Center, De Boelelaan 1117, 1081 HV Amsterdam, The Netherlands
b
Epilepsy Center Kempenhaeghe, Sterkselseweg 65, 5591 VE Heeze, The Netherlands
abstractarticle info
Article history:
Received 27 April 2011
Revised 7 July 2011
Accepted 8 July 2011
Available online 23 July 2011
Keywords:
EEG/fMRI
Electrode positions
Point matching
Hungarian algorithm
The analysisof simultaneous EEG and fMRI datais generally based on the extraction of regressorsof interestfrom
the EEG, whichare correlated to the fMRI datain a generallinear model setting.In more advanced approaches, the
spatial information of EEG is also exploited by assuming underlying dipole models. In this study, we present a
semi automatic and efficient methodto determine electrode positions fromelectrode gel artifacts, facilitatingthe
integration of EEG and fMRI in future EEG/fMRI data models.
In order to visualize all electrode artifacts simultaneously in a single view, a surface rendering of the structural
MRI is made using a skin triangular mesh model as reference surface, which is expanded to a “pancake view”.
Then the electrodes are determined with a simple mouse click for each electrode. Using the geometry of the skin
surface and its transformation to the pancake view, the 3D coordinates of the electrodes are reconstructed in the
MRI coordinate frame.
The electrode labels are attached to the electrodepositions by fittinga templategrid of the electrode cap in which
the labels are known. The correspondence problem between template and sample electrodes is solved by
minimizing a cost function over rotations, shifts and scalings of the template grid. The crucial step here is to use
the solution of the so-called “Hungarian algorithm”as a cost function, which makes it possible to identify the
electrode artifacts in arbitrary order. The template electrode grid has to be constructed only once for each cap
configuration.
In our implementation of this method,the whole procedure can be performed within 15 min including import of
MRI, surface reconstruction and transformation, electrode identification and fitting to template. The method is
robust in the sense that an electrode template created for one subject can be used without identification errors
for another subject for whom the same EEG cap was used. Furthermore, the method appears to be robust against
spurious or missing artifacts. We therefore consider the proposed method as a useful and reliable tool within the
larger toolbox required for the analysis of co-registered EEG/fMRI data.
© 2011 Elsevier Inc. All rights reserved.
Introduction
Electroencephalography (EEG) represents time varying potential
differences that can be recorded at electrodesensors on the humanscalp
and is caused by synchronous synaptic currents associated with
interacting neurons. EEG signals have a very high temporal resolution,
comparable to the time scale of the recorded brain processes. The most
important limitation of EEG is its poor spatial resolution, caused by
mixing of scalp potentials due to neighboring dendritic currents. Source
localization with EEG is based on the construction of mathematical
models in which the spatiotemporal distribution of the surface
potentials is predicted in terms of assumed current dipole sources.
Functional Magnetic Resonance Imaging (MRI) on the other hand,
which yields a series of three dimensional T2* images (Buxton, 2002)
representing inhomogeneity in magnetic susceptibility, has a high
spatial resolution and a temporal resolution which is much lower than
that of the working brain. Contrary to EEG, which is a manifestation of
Ohm's law, the physical meaning of fMRI signals is much more
ambiguous. Neurovascular coupling converts neural activity into local
changes in blood volume, blood flow, and blood oxygenation (the
BOLD effect), and causes measurable variations in T2*.
Since EEG and fMRI are complementary, several studies have
appeared in the literature where the relative strengths of both
techniques are combined using co-registered EEG and fMRI (EEG/
fMRI) (e.g. De Munck et al., 2007, 2009; Goldman et al., 2002; Laufs et al.,
2003; Salek-Haddadi et al., 2003). Clinical applications of EEG/fMRI are
mainly in the pre-surgical evaluation of patients with epilepsy (e.g.
Bagshaw et al., 2005; Benar et al., 2006; Jacobs et al., 2008; Thornton
et al., 2010; Zijlmans et al., 2007). The methodology by which EEG and
fMRI are combined is generally based on EEG-informed fMRI analysis
(Ullsperger, 2010) which implies that EEG events or amplitude
variations of certain rhythms are used in a general linear model (GLM)
NeuroImage 59 (2012) 399–403
⁎Corresponding author. Fax: +31204444147.
E-mail address: jc.munck@vumc.nl (J.C. de Munck).
1053-8119/$ –see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.neuroimage.2011.07.021
Contents lists available at ScienceDirect
NeuroImage
journal homepage: www.elsevier.com/locate/ynimg
Author's personal copy
with fMRI time series as explained variables. Current practice is that
these EEG-regressors are either derived from channels where EEG events
are most prominent or result from a pre-processing step such as
independent component analysis (ICA), see for example Eichele et al.
(2005) or Debener et al. (2010).
Important limitations of this approach are that the spatial
information present in the EEG is not fully exploited and that the
surface potentials are treated as substitutes for local neural currents.
Although with the ICA approach these objections are met to some
degree, the integration of EEG and fMRI would substantially benefit
from the development of a new generation of models in which both
EEG and fMRI are treated as dependent variables and in which neural
sources are treated as unknowns (Daunizeau et al., 2010; De Martino
et al., 2010). Apart from the more theoretical aspects that need to be
addressed for the proper development of such models, a more
practical aspect is the determination of the electrode positions with
respect to the anatomical MRI, such that the EEG potential differences
can be accurately predicted using dipole models (e.g. De Munck et al.,
1988, 1991). The electrode localization problem also arises when
regions of significant EEG/fMRI-correlation are validated using dipolar
models (e.g. Lemieux et al., 2001). Finally, the authors foresee that
data driven approaches to integrate EEG and fMRI, like ICA and group
ICA (Calhoun et al., 2009), could benefit from the incorporation of
electrode positions from which the EEG is recorded.
The main contribution of this technical note is to present a simple,
efficient and novel method to determine EEG electrode positions
using the small artifacts that the electrode gel imprints on the
anatomical MRI scan. As an advantage compared to the use of 3D
digitizer devices, artificial electrode markers (Sijbers et al., 2000)or
laser scanners (Koessler et al., 2011), no additional preparation time
with the patient is required and the electrode positions are obtained
directly in MRI coordinates. Therefore, no accuracy is lost with
additional transformation from pointer device to MRI space. Further-
more, the method can be applied retrospectively to existing EEG/fMRI
data sets where the artifacts are sufficiently visible. Finally, when the
algorithm is fully integrated with the method of Van't Ent et al.
(2001), our proposed method automatically yields a boundary
element model (BEM) for the EEG forward computations.
Methods
Subjects and data
The method was tested using the data acquired in an earlier EEG/
fMRI study (De Munck et al., 2007). EEG was acquired using an MR
compatible EEG amplifier (SD MRI 64, MicroMed, Treviso, Italy) and a
nylon cap providing 62 Ag/AgCl electrodes positioned according to the
extended 10–20 system and two additional ground electrodes.
Electrodes are attached to small cups with inner diameter of 10 mm
and 4 mm in height, inserted in the cap and filled with Electro Gel™to
minimize the contact impedance. Of the 17 healthy subjects 12 wore a
small (circumference 50–54 cm) and 5 wore a large (circumference
54–58 cm) MRI-compatible electrode cap of MicroMed. Data of the 12
subjects wearing the small cap were used to test the automatic
labeling and the data of the other 5 subjects were used to evaluate the
proposed method in a practical setting. Anatomical MRI scans were
made on a 1.5 T Siemens Sonata using a T1-weighted sequence
(MPRAGE, TR=2700 ms, T1 = 950 ms, TE= 5.18 ms, 256×192 × 160
matrix, FOV =256 ×192 ×240 mm, voxel size =1.0 ×1.0 ×1.5 mm
3
)
consisting of 160 coronal slices of 1.5 mm in thickness.
Visualization of artifacts
Fig. 1 shows a sagittal slice of an MRI scan on which a rough
delineation (yellow line) of the head contour was drawn. The yellow
dots indicate the cross sections of outlines of the head delineated in
the other two orthogonal directions. The red contour is the result of a
skin surface fit (a triangular mesh) obtained through the method of
Van't Ent et al. (2001). Only a few head outlines suffice to create
approximations of the head with an accuracy of a few mm. These head
outlines can also be used to predict brain and skull compartments for
BEM modeling applications Van't Ent et al. (2001). The white arrows
indicate a few electrode artifacts that are located near the chosen slice.
An overview of all electrode artifacts can be obtained using the
spherical parameterization r(θ,φ) underlying Van 't Ent's surface
model. Here ris the distance from the center of the head, θis the angle
w.r.t. the vertical axis and φis the angle in the horizontal plane. The
“pancake views”in Fig. 2 are constructed by integrating the MRI pixel
values from r(θ,φ)−δ
1
to r(θ,φ)+δ
2
and plotting the result using θas
radial and φas angular coordinate. The integration of pixel values
makes the pancake view robust for deviations of the head surface
from the MRI scan. It appears that δ
1
=5 mm and δ
2
=10 mm are
appropriate choices. The level and window can be adjusted such that
all electrode artifacts become visible in a single view.
Template grid
Electrode positions are determined by double clicking on the
pancake view. Each mouse click corresponds to a θand φcoordinate,
and using the surface model r(θ,φ) this yields the 3D positions in
spherical coordinates (r,θ,φ). Since the origin of these spherical
coordinates is fixed to the center of the head surface and is known in
MRI coordinates, this semi-automatic tool is able to provide a list of 3D
electrode positions in MRI space. In our implementation of the
algorithm mistakes can easily be adjusted or deleted and new points
can be added to an existing point list imported from disk. For dipolar
modeling, appropriate electrode labels need to be attached to each of
the electrode positions. Therefore, a procedure has been implemented
to attach an electrode label to each 3D position and hence to create a
template electrode grid. Although this is quite time consuming, the
Fig. 1. A sagittal view is presented of the surface fitting tool. The yellow contour is a very
rough outline of the skin. The yellow dots represent cross sections of similar outlines in
the axial and coronal direction. The red contour is the result of the head surface fit. The
white arrows indicate where on this slice a few of the electrode artifacts can be
observed.
400 J.C. de Munck et al. / NeuroImage 59 (2012) 399–403
Author's personal copy
creation of the template only needs to be performed once for each cap
and can then be used for other subjects wearing the same cap.
Finding electrode labels
The underlying idea of an automated method to find the electrode
labels for the “clicked points”, here called sample points, is to match
these points to a template made from another subject wearing the
same electrode cap and to pick out the label of the closest template
point. This point matching problem is impeded, because the electrode
positions could be indicated in arbitrary order. Therefore, one cannot
use a simple point-to-point matching algorithm and apart from the
matching parameters, one also has to optimize the permutation
template points such that the total sum of distances between pairs of
electrodes x
i
and template t
k
is as small as possible. When the number
of sample electrodes and the number of template points are equal, the
problem can be formulated mathematically as:
Cost = min
T
min
π∑
i
xi−TtπiðÞ
ð1Þ
where the transformation Tis a combination of shifts, rotations and
scalings (9 parameters)
T
x
y
z
0
@1
A
0
@1
A=
λx00
0λy0
00λz
0
@1
A
rxx rxy rxz
ryx ryy ryz
rzx rzy rzz
0
@1
A
x
y
z
0
@1
A+
tx
ty
tz
0
@1
A;ð2Þ
and π() is a permutation of the electrodes. The permutation for which
the cost defined in Eq. (1) is minimal can be used to find the optimal
label of electrode iby taking label π(i) from the template.
For a given transformation T, the optimal permutation problem
can be considered as an optimal assignment problem. For that
purpose the matrix M
ik
is defined, with
Mik =xi−Ttk
ðÞjj:ð3Þ
The problem of picking exactly one unique row for every column of
M
ik
such that the sum of all picked elements is as small as possible, is
well studied in the field of econometrics and network theory and can
be solved using the so-called Hungarian algorithm. This name was
given to the algorithm by Kuhn (1955), who first studied it in detail
and it was further optimized by Munkres (1957), who observed that
the problem could be solved in O(K
4
) amount of time, where Kis the
number of columns. There are several free implementations available
and we used one based on the source code of Stachniss (2006), who
wrote a C-version with a computational complexity of O(K
3
).
When spurious electrodes are clicked, or electrodes are missing, this
would result in a non-square matrix in Eq. (3). Following Stachniss
(2006) we solve this problem by adding zero rows or columns to M
ik
to
make it square. Electrodes assigned added null columns and null rows
are interpreted as spurious and missing, respectively. The optimal
transformation Twas found using Nelder–Mead's simplex algorithm
(Press et al., 2007) with the solution of the assignment problem as cost
function. Therefore, the assignment problem had to be solved typically
several hundreds of times, with different transformations T.Thescaling
parameters were initialized by unity and the rotation and shift were
initialized using the optimal match determined from first, second and
third moments of the template and sample electrodes (see e.g. Jakličand
Solina, 2003). Fig. 3 demonstrates how the label matching algorithm
works in practice.
Validation of the labeling algorithm
Electrode positions were determined for all 12 subjects wearing the
small cap, using the mouse click tool described above. The true labels of
all electrodes in each electrode grid were determined by applying the
manual labeling procedure. The automatic labeling algorithm was
tested using a series of simulations in which the electrodes of one
subject were treated as sample electrodes and the labeled electrodes of
another subject were used as template. To study the robustness of the
labeling algorithm we determined a histogram of the number of
mislabeled electrodes over all 11×12 = 132 combinations of electrode
grids. The nine transformation parameters resulting from the fit
procedure were ignored. In additional simulations, either a spurious
electrodewas addedto sample electrodes or onewas removed. Since in
this type of simulations the sample and template grids were always
different, error histograms were made over each of the all 144
combinations of electrode grid combinations.
Results
The simulations where each electrode grid was fit to each of the
others, resulted in 129 cases (97.8%) where no electrodes obtained the
wrong label, in 2 cases (1.5%) where two electrodes were wrong and 1
case (0.75%) where more than 50 electrodes wrong. The latter case
resulted in a final cost of about 1 cm, which is more than twice as large
as the final cost in the more successful cases.
Three simulations were performed by adding an additional
electrode to the sample electrodes: an electrode halfway Fz and Fcz,
or one halfway AF4 and AF8, or one halfway FT8 and F8. Resulting
error histograms were averaged and the result is presented in Table 1
(second column). The presence of a spurious sample electrode, did, in
85% of the cases, not result in any labeling error. In 10% of the cases
there were more than 10 falsely labeled electrodes. The third column
of Table 1 shows the effect of removing an electrode, averaged over all
possible electrode removals and pairs of electrode grids. It appears
that in more than 93% of the cases this does not have any impact on
the correctness of the labeling algorithm.
The algorithms used are implemented in Brain Image Analysis
Package (http://demunck.info/software/). Using this implementation,
the whole procedure could be performed within 15 min including
import of the anatomical MRI, surface construction and transformation,
Fig. 2. The (averaged) pancake view of the skin clearly shows all electrode artifacts
simultaneously. Each mouse click on an electrode artifact is converted to a θand
φ- coordinates, and, combined with the surface model r(θ,φ), this yields the 3D
Cartesian coordinates of the electrodes.
401J.C. de Munck et al. / NeuroImage 59 (2012) 399–403
Author's personal copy
electrode identification and fitting to template as was determined using
the MRI data of the five subjects wearing the larger cap.
Discussion and conclusion
During measurement setup no special attention waspaid to a precise
placement of the cap and therefore we consider the simulation
condition in which each electrode grid is matched to each of the others
as an unfavorable. We also expect that more accurate templates could
have been obtained by averaging sampled electrode grids from many
different subjects. Nevertheless, our proposed method appears to be
quite robust. The only case where more than half of the electrodes
receive a wrong label was easily recognized by visual inspection and by
the large remaining residual error.
The absolute accuracy by which the electrode positions are obtained
is hard to quantify precisely because no gold standard is available.
However, based on the inner radius of the electrode cups of 10 mm and
the possibility of the human observer to use the symmetry of the cap
configuration, we expect that it is well possible to determine the center
of gravity of the electrode artifacts with an accuracy of about 3 mm on
average. Based on simulation studies by De Munck et al. (1991), Khosla
et al. (1999) and Wangand Gotman (2001), this would contributeabout
a few mm to the dipole localization error, dependent on the precise
choice of the inverse model. We would like to emphasize that with our
method, the electrode positions are obtained directly in MRI co-
ordinates. Other methods, based on additional hardware to digitize
points (e.g. Koessler et al., 2011), also require additional coordinate
matching, which further compromises the electrode position accuracy.
The proposed method was also tested using other hardware than
presented in the methods section. This is important because the
presence of an EEG cap during structural MRI scanning causes several
types of scanning artifacts, implying that apart from the electrode gel
also the wires cause imaging artifacts, as shown by Mullinger et al.
(2008). As an example, we present in Fig. 4 the pancake view of a more
modern Micromed 61 channel cap, with smaller cups (inner diameter
5 mm, height 3 mm) made of much softer material, using a different
type of electrode gel (Plastic Gel for EEG, Paste MT10) scanned with a
Table 1
Labeling error distribution caused by a spurious or missing electrode of the sampled
electrode grid.
N error % of the cases
Spurious electrode Missing electrode
0 84.5 93.1
1 4.17 0.57
2 4.4 2.5
3 0.93 0.20
4 0. 0.03
5 0. 0.02
N10 6.0 3.61
Fig. 4. A pancake view of the electrode artifacts is shown using different hardware than
used in Figs. (1), (2) and (3). The inner diameter of the cups is smaller (5 mm), the cup
material is softer, a Philips 3T scanner was used and electrode gel was different.
Fig. 3. The yellow dots on the pancake view in panel A represent the electrode positions
obtained from mouse clicks on the screen. The (labeled) red dots indicate the electrodes
from the template, projected onto the pancake view. After the matching procedure,
panel B, the clicked points (now presented in blue) obtain the best possible
permutation of labels from the template.
402 J.C. de Munck et al. / NeuroImage 59 (2012) 399–403
Author's personal copy
Philips 3T scanner. It appears that the electrode artifact visibility is
comparable to that obtained with the larger electrodes on a Siemens
scanner using Electro Gel™.
Based on these findings and practical experience, we consider our
proposed method a valuable practical and robust tool for the further
integration of EEG and fMRI data.
References
Bagshaw, A.P., Hawco, C., Benar, C.G., Kobayashi, E., Aghakhani, Y., Dubeau, F., Pike, G.B.,
Gotman, J., 2005. Analysis of the EEG-fMRI response to prolonged bursts of
interictal epileptiform activity. NeuroImage 24, 1099–1112.
Benar, C.G., Grova, C., Kobayashi, E., Bagshaw, A.P., Aghakhani, Y., Dubeau, F., Gotman, J.,
2006. EEG-fMRI of epileptic spikes: concordance with EEG source localization and
intracranial EEG. NeuroImage 30, 1161–1170.
Buxton, R.B., 2002. Introduction to Functional Magnetic Resonance Imaging. Principles
and Techniques. Cambridge University Press, Cambridge.
Calhoun, V.D., Liu, J., Adali, T., 2009. A review of group ICA for fMRI data and ICA for joint
inference of imaging, genetic, and ERP data. NeuroImage 45, S163–S172.
Daunizeau, J., Laufs, H., Friston, K.J., 2010. EEG-fMRI information fusion: biophysics and
data analysis. In: Mulert, C., Lemieux, L. (Eds.), EEG-fMRI: Physiological Basis,
Technique, and Applications. Springer, pp. 511–526.
De Martino, F., Valente, G., De Borst, A.W., Esposito, F., Roebroeck, A., Goebel, R.,
Formisano, E., 2010. Multimodal imaging: an evaluation of univariate and
multivariate methods for simultaneous EEG/fMRI. MRI 28, 1104–1112.
De Munck, J.C., Van Dijk, B.W., Spekreijse, H., 1988. Mathematical dipoles are adequate
to describe realistic generators of human brain activity. IEEE Trans. Biomed. Eng.,
BME 35 (11), 960–966.
De Munck, J.C., Vijn, P.C.M., Spekreijse, H., 1991. A method to determine electrode
positions with respect to the head. Electroenceph. Clin. Neurophysiol. 78, 85–87.
De Munck, J.C., Gonçalves, S.I., Huijboom, L., Kuijer, J.P.A., Pouwels, P.J.W., Heethaar,
R.M., Lopes da Silva, F.H., 2007. The hemodynamic response of the alpha rhythm:
an EEG/fMRI study. NeuroImage 35 (3), 1142–1151.
De Munck, J.C., Gonçalves, S.I., Mammoliti, R., Heethaar, R.M., Lopes da Silva, F.H., 2009.
Interactions between different EEG frequency bands on alpha-fMRI correlations.
NeuroImage 47, 69–76.
Debener, S., Thorne, J., Schneider, T.R., Campos Viola, F., 2010. Using ICA for the analysis
of multi-channel EEG data. In: Ullsperger, M., Debener, S. (Eds.), Simultaneous EEG
and fMRI: Recording, Analysis and Application. Oxford University Press, Oxford, pp.
121–133.
Eichele, T., Specht, K., Moosmann, M., Jongsma, M.L., Quiroga, R.Q., Nordby, H., Hugdahl,
K., 2005. Assessing the spatiotemporal evolution of neuronal activation with single-
trial event-related potentials and functional MRI. PNAS 102 (49), 17798–17803.
Goldman, R.I., Stern, J.M., Engel Jr., J., Cohen, M.S., 2002. Simultaneous EEG and fMRI of
the alpha rhythm. Neuroreport 13 (18), 2487–2492.
Jacobs, J., Hawco, C., Kobayashi, E., Boor, R., LeVan, P., Stephani, U., Siniatchkin, M.,
Gotman, J., 2008. Variability of the hemodynamic response as a function of age and
frequency of epileptic discharge in children with epilepsy. NeuroImage 40,
601–614.
Jaklič, A., Solina, F., 2003. Moments of superellipsoids and their application to range
image registration. IEEE Trans. Sys. Man Cyb., Part B 33 (4), 648–657.
Khosla, D., Don, M., Kwong, B., 1999. Spatial mislocalization of EEG electrodes —effects
on accuracy of dipole estimation. Clin. Neuroph. 110 (2), 261–271.
Koessler, L., Cecchin, T., Caspary, O., Benhadid, A., Vespignani, H., Maillard, L., 2011.
EEG–MRI co-registration and sensor labelling using a 3D laser scanner. Ann.
Biomed. Eng. 39 (3), 983–995.
Kuhn, H.W., 1955. The Hungarian method for the assignment problem. Nav. Res. Logist.
Quart. 2, 83–97.
Laufs, H., Kleinschmidt, A., Beyerle, A., Eger, E., Salek-Haddadi, A., Preibisch, C., Krakow,
K., 2003. EEG-correlated fMRI of human alpha activity. NeuroImage 19 (4),
1463–1476.
Lemieux, L., Krakow, K., Fish, D.R., 2001. Comparison of spike-triggered functional MRI
BOLD activation and EEG dipole model localization. NeuroImage 14 (5), 1097–1104.
Mullinger, K., Debener, S., Coxon, R., Bowtell, R., 2008. Effects of simultaneous EEG
recording on MRI data quality at 1.5, 3 and 7 tesla. Int. J. Psychophysiol. 67 (3),
178–188.
Munkres, J., 1957. Algorithms for the assignment and transportation problems. J. Soc.
Indust. Appl. Math. 5 (1), 32–38.
Press, W.H., Flannery, B.P., Teukolsky, S.A., Wetterling, W.T., 2007. In: Cambridge
University Press (Ed.), Numerical Recipes in C++, Third edition, pp. 82–93.
Salek-Haddadi, A., Friston, K.J., Lemieux, L., Fish, D.R., 2003. Studying spontaneous EEG
activity with fMRI. Brain Res. Brain Res. Rev. 43 (1), 110–133.
Sijbers, J., Vanrumste, B., Van Hoey, G., Boon, P., Verhoye, M., Van der Linden, A.-M., Van
Dyck, D., 2000. Automatic localization of EEG electrode markers within 3D MR data.
MRM 18 (4), 485–488.
Stachniss C., 2006, Exploration and Mapping with Mobile Robots, PhD-thesis,
University of Freiburg, Department of Computer Science.
Thornton, R., Laufs, H., Rodionov, R., Cannadathu, S., Carmichael, D.W., Vulliemoz, S.,
Salek-Haddadi, A., McEvoy, A.W., Smith, S.M., Lhatoo, S., Elwes, R.D., Guye, M.,
Walker, M.C., Lemieux, L., Duncan, J.S., 2010. EEG correlated functional MRI and
postoperative outcome in focal epilepsy. J Neurol. Neurosurg. Psychiatry 81,
922–927.
Ullsperger, M., 2010. EEG-informed fMRI analysis, chapter 3.3. In: Ullsperger, M.,
Debener, S. (Eds.), Simultaneous EEG and fMRI: Recording, Analysis and
Application. Oxford University Press, pp. 153–159.
Van't Ent, D., De Munck, J.C., Kaas, A.L., 2001. An automated procedure for deriving
realistic volume conductor models for MEG/EEG source localization. IEEE Trans.
Biomed. Eng., BME 48 (12), 1434–1443.
Wang, Y., Gotman, J., 2001. The influence of electrode location errors on EEG dipole
source localization with a realistic head model. Clin. Neuroph. 112 (9), 1777–1780.
Zijlmans, M., Huiskamp, G., Hersevoort, M., Seppenwoolde, J.H., Van Huffelen, A.C.,
Leijten, F.S., 2007. EEG-fMRI in the preoperative work-up for epilepsy surgery.
Brain 130, 2343–2353.
403J.C. de Munck et al. / NeuroImage 59 (2012) 399–403