REVIEW

published: 04 April 2019

doi: 10.3389/fneur.2019.00325

EEG Source Imaging: A Practical

Review of the Analysis Steps
Christoph M. Michel 1,2* and Denis Brunet 1,2
1
Department of Basic Neurosciences, University of Geneva, Geneva, Switzerland, 2 Center for Biomedical
Imaging Lausanne-Geneva (CIBM), Geneva, Switzerland

The electroencephalogram (EEG) is one of the oldest technologies to measure neuronal

activity of the human brain. It has its undisputed value in clinical diagnosis, particularly
(but not exclusively) in the identification of epilepsy and sleep disorders and in the
evaluation of dysfunctions in sensory transmission pathways. With the advancement
of digital technologies, the analysis of EEG has moved from pure visual inspection of
amplitude and frequency modulations over time to a comprehensive exploration of the
temporal and spatial characteristics of the recorded signals. Today, EEG is accepted as a
powerful tool to capture brain function with the unique advantage of measuring neuronal
processes in the time frame in which these processes occur, namely in the sub-second
range. However, it is generally stated that EEG suffers from a poor spatial resolution that
makes it difficult to infer to the location of the brain areas generating the neuronal activity
measured on the scalp. This statement has challenged a whole community of biomedical
engineers to offer solutions to localize more precisely and more reliably the generators
Edited by: of the EEG activity. High-density EEG systems combined with precise information of the
Sandor Beniczky,
head anatomy and sophisticated source localization algorithms now exist that convert
Aarhus University Hospital, Denmark
the EEG to a true neuroimaging modality. With these tools in hand and with the fact
Reviewed by:
Bruno J. Weder, that EEG still remains versatile, inexpensive and portable, electrical neuroimaging has
University of Bern, Switzerland become a widely used technology to study the functions of the pathological and healthy
Jurgis Strautmanis,
Riga Stradinš University, Latvia
human brain. However, several steps are needed to pass from the recording of the EEG to
*Correspondence:
3-dimensional images of neuronal activity. This review explains these different steps and
Christoph M. Michel illustrates them in a comprehensive analysis pipeline integrated in a stand-alone freely
christoph.michel@unige.ch
available academic software: Cartool. The information about how the different steps are
Specialty section:
performed in Cartool is only meant as a suggestion. Other EEG source imaging software
This article was submitted to may apply similar or different approaches to the different steps.
Applied Neuroimaging,
a section of the journal Keywords: EEG, pre-processing, source localization, head model, inverse model
Frontiers in Neurology

Received: 05 January 2019

Accepted: 15 March 2019
INTRODUCTION
Published: 04 April 2019
The electric potential differences between electrodes placed on distinct scalp positions is due to
Citation:
the propagation of current flow induced by synchronized post-synaptic potentials of pyramidal
Michel CM and Brunet D (2019) EEG
Source Imaging: A Practical Review of
neurons in the head according to Poisson’s equations (1). However, this propagation is not
the Analysis Steps. homogenous. The current flow is strongly attenuated by the skull due to its high resistivity.
Front. Neurol. 10:325. This attenuation has to be properly modeled when solving the so-called forward problem, i.e.,
doi: 10.3389/fneur.2019.00325 determining the potential at each scalp electrode generated by a known source in the brain (2).

Frontiers in Neurology | www.frontiersin.org 1 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

Since the thickness of the skull is not homogeneous across the make a priori assumption with respect to the number of
head, it is highly recommended that the individual anatomical dipoles. The most popular distributed source models currently
information derived from the MRI is used to determine the used in the EEG community are modifications of a solution
skull thickness and thus the local conductivity properties. Also, initially proposed by Hämäläinen and Ilmoniemi (18), called
the shape of the head is not spherical and thus the distance the Minimum Norm Solution (MN). The constraint introduced
of the electrodes to the center of the head is variable. The in this solution is that the current distribution over all
exact position of each electrode on the individual head should solution points has minimum energy (minimizing the least-
therefore be known. These properties (local skull thickness and square error, i.e., the L2-norm) and that the forward solution
3D electrode position) are then incorporated in the lead field, of this distribution optimally explains the measured data. MN
which determines how the electric activity at a certain electrode solutions are biased toward superficial sources because of their
is related to the activity of the different sources in the brain. spatial vicinity to the sensors. Therefore, weighting parameters
The more precise and anatomically correct this lead field is have been introduced to mitigate this bias, leading to the so-
determined, the more precise the source localization will be (3). called weighted minimum norm (WMN) solutions (19–21).
Once the proper head model has been built and the lead A variation of WMN is the low resolution electromagnetic
field is constructed, the second step consists in solving the tomography (LORETA) in which the norm of the second-
inverse problem, i.e., determining the intracranial sources that order spatial derivative of the current source distribution is
generate a given EEG potential measurement on the scalp. This minimized to ensure spatial coherence and smoothness (22).
inverse problem is a fundamental challenge because a very large This constraint has been justified by the physiological plausible
number of different source distributions can produce the same assumption that activity in neighbored voxels are correlated.
potential field on the scalp (4). Due to this non-uniqueness, a Another modification has been suggested by Grave de Peralta
priori assumptions need to be incorporated (5). They can be Menendez (23), called LAURA (Local AUtoRegressive Average).
purely mathematical or include neurophysiological, biophysical It incorporates the biophysical law that the strength of the source
and anatomical knowledge about the distribution of neuronal falls off with the inverse of the cubic distance for vector fields.
activity in space and in time. It must be made very clear that no LAURA integrates this law in terms of a local autoregressive
matter how sophisticated these assumptions and constraints are, average with coefficients depending on the distances between
the provided source solution remains an estimation that depends solution points. The general communality of all these linear
on how well-genuine sources conform to these assumptions (6). inverse solutions is that they provide a distribution of the current
This holds for the EEG as well as for the MEG. density in the whole brain volume that is described as a 3D grid of
Localization of a limited number of equivalent dipoles is discrete solution points. In each of these solution points, a current
the most classical approach to solve the inverse problem (7). dipole with a certain orientation and strength is estimated.
The a priori assumption in this solution is that only one or a Usually, the space of these solution points is restricted to the
few active areas in the brain generated the scalp potential field. gray matter (24). Several other linear and non-linear source
Under this constraint, non-linear multidimensional optimization localization algorithms have been described in the literature. This
procedures allow to determine the dipole parameters that best review focuses on the pre-processing steps that are needed for
explain the observed scalp potential measurements in a least- source localization and not on the characteristics of the different
square sense (8, 9). The maximal number of dipoles that can inverse solutions. For detailed discussions we refer to previous
be reliably localized depends on the number of scalp electrodes comprehensive review articles (3, 25–28).
and is further limited by the non-linear complexity of the search In the following, we describe the different steps that are
algorithms with multiple sources (5). The number of dipoles needed to get to these source localizations by illustrating
can be increased by searching for the best solutions of dipoles them with the implementation in our freely available academic
with time-varying strength over a certain time period and by software package Cartool, a stand-alone program for the spatio-
decoupling the linear and non-linear part of the estimation temporal analysis of EEG and evoked potentials (29), https://
[BESA, (10), MUSIC, (11)]. It is important to be aware of sites.google.com/site/cartoolcommunity/. The purpose of this
the fact that if the number of dipoles is underestimated the concrete illustration is to explain in detail the points that
source localization is biased by the missing dipoles. On the are important to consider in each processing step and how
other hand, if too many dipoles are assumed, spurious sources they are implemented in Cartool. Several other powerful
will be introduced. Nevertheless, dipole source localization can commercial or academic software packages for EEG source
produce reasonable results under some particular conditions imaging exist that have implemented similar or alternative
(12), in particular in localizing the irritative zone in focal epilepsy strategies. A comprehensive overview of different academic
(13–15) or the localization of primary sensory areas in evoked software applications can be found in a special issue of the Journal
potentials, such as the sensorimotor cortex in surgical candidates Computational Intelligence and Neuroscience (30), where
(16). Dipole source localization is still widely used in the MEG programs such as BrainStorm (31), EEGLAB (32), FieldTrip (33),
community for these clinical applications (17). NUTMEG (34), SPM (35), and Cartool (29) are described. Widely
Recent development in brain source imaging has offered more used commercially available software packages for EEG/MEG
exciting options to localize brain sources from scalp EEG signals source localization are BESA, Curry, GeoSource, and BrainVision
and have largely replaced the dipole source localization approach. Analyzer. Table 1 gives a summary of some of the most often
These so-called distributed source localization methods do not used software packages and the source localization methods that

Frontiers in Neurology | www.frontiersin.org 2 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

TABLE 1 | Non-exhaustive list of academic and commercial software packages that offer EEG source localization tools.

Name Website Inverse models

ACADEMIC SOFTWARE PACKAGES

Brainstorm https://neuroimage.usc.edu/brainstorm Dipole modeling, Beamformer, sLORETA, dSPM
Cartool https://sites.google.com/site/ Minimum Norm, LORETA, LAURA, Epifocus
cartoolcommunity/
EEGLab https://sccn.ucsd.edu/eeglab/index.php Dipole modeling
Fieldtrip http://www.fieldtriptoolbox.org/ Dipole modeling, Beamformer, Minimum Norm
LORETA http://www.uzh.ch/keyinst/loreta.htm LORETA, sLORETA, eLORETA
MNE https://martinos.org/mne/stable/index.html MNE, dSPM, sLORETA, eLORETA
NUTMEG https://www.nitrc.org/projects/nutmeg Beamformer
SPM https://www.fil.ion.ucl.ac.uk/spm/ dSMP
COMMERCIAL SOFTWARE PACKAGES
BESA http://www.besa.de/products/besa-research/ Dipole modeling, RAP-MUSIC, LORETA, sLORETA,
besa-research-overview/ LAURA, SSLOFO
brainvision analyzer https://www.brainproducts.com/ LORETA
BrainVoyager https://www.brainvoyager.com/ Beamformer, Minimum Norm, LORETA, LAURA
GeoSource https://www.usa.philips.com/healthcare/ Minimum Norm, LORETA, sLORETA, LAURA
solutions/neuro/neuro-research-applications
CURRY https://compumedicsneuroscan.com/curry- Dipole modeling, MUSIC, Beamformer, Minimum
source-reconstruction/ norm, sLORETA, eLORETA, SWARM

they implemented. Whatever software is used, it is crucial that and an increasing recognition of physiological relevance of
the user is aware and informed about the implementation of frequencies below and above the conventional EEG frequencies
the different processing steps. In view of recent efforts to setup [infraslow frequencies in resting state activity (37), high
best practice guidelines of reporting EEG/MEG studies (https:// frequency oscillations in normal and pathological brains (38)],
cobidasmeeg.wordpress.com/), having access to the information the range of the band-pass filter is driven by the study question.
of how the steps are done in the different software packages and Resting-sate EEG is often filtered between 1–40 Hz, while evoked
reporting this information in the publications is important to potential data usually considers broader frequency ranges (0.1–
ensure reproducibility and replicability. 100 Hz). Filtering the data can have important effects on the
time-courses and the phases of the data (39, 40), as well as
BASIC REQUIREMENTS on the localization of the waveforms’ local extrema. This is of
particular relevance in evoked potential studies, time-frequency
EEG Pre-processing analysis and connectivity measures. The exact characteristics of
Raw EEG data are contaminated by artifacts from many the filter that has been used should be described in the study
non-physiological (power line, bad electrode contact, broken report (36). In Cartool, we implemented a non-causal, Infinite
electrodes, etc.) and physiological (cardiac pulse, muscle activity, Impulse Response (IIR) Butterworth filter of 2nd order, known
sweating, movement, etc.) sources. These artifacts have to be for its optimally flat passband response, which limits the artificial
carefully identified and either removed or excluded from further introduction of new local maxima (41). Both Butterworth low-
analysis. This is a cumbersome work and should be done by visual and high-pass filters have a−12 db/octave roll-off, and are
inspection of the raw data by experienced electrophysiologists. computed linearly with forward and backward passes, which
However, with the increasing availability of public EEG databases eliminates any phase shifts. This ensures that the local maxima
and the desire to analyze large datasets, the need for and the will remain at their expected positions, irrespectively of their
usage of automatic artifact detection and removal software is frequency content. In the specific case of Butterworth high-pass
on the rise. Blindly applying such programs is problematic, filtering, the D.C. value is explicitly removed beforehand, as very
because the type of artifacts is manifold and can vary in different high baselines could cause IIR filters to become instable.
experimental conditions. It is therefore recommended that if
automatic artifact detection and correction methods are used, Down-Sampling
they should still be followed up by visual inspection of the data After filtering, it is often useful to down-sample the data
(36). In the following we describe the pre-processing pipeline as most of the frequencies higher than the low-pass cutting
implemented in Cartool. frequency should be gone. It can dramatically reduce the memory
requirements for the subsequent processing, without losing
Temporal Filtering any information. The Nyquist theorem would require down-
Most studies first apply a temporal filter to the data in sampling not lower than twice the highest remaining frequency.
order to remove frequencies that are considered to be non- In practice, though, because the filters’ cut-offs are never perfectly
physiological and/or non-relevant for the study at hand. Since sharp, and in order to keep some additional time resolution,
there is no consensus regarding the relevant frequency range the final sampling frequency should be chosen to be about four

Frontiers in Neurology | www.frontiersin.org 3 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

times the highest remaining frequency after low-pass filtering. In This is very similar to an Interquartile Mean (IQM), but cutting
Cartool and for integer down-sampling ratios, down-sampling is the Cumulated Density Function into 7 intervals instead of 4, so
done with a Cascaded Integrator-Comb (CIC) filter (42), which we technically have an Inter Septile Weighted Mean. For each
in practice is quite easy to compute in off-line applications. electrode e:
Other software packages, such as for example EEGLAB (32) apply X  X 
antializing filters to reduce the sampling frequency. i=5 vi i=5 1
SpatialFilter(e) = / (1)
i=1 di i=1 di

Electrode Interpolation and ICA With vi being the 5 remaining voltage values from the 6 nearest
In Cartool, data inspection is performed semi-automatically. neighbors of electrode e, plus the central value, each being at
The user scrolls through the data and the program detects and distance di . An example of the effect of the spatial filter on
visualizes electrodes with amplitudes above a certain range. If the waveforms (Figures 1B,C), but more importantly on the
the user decides that a given electrode is an outlier due to bad topography, can be seen in Figures 1D,E.
contact, this electrode is marked and ignored in the subsequent
independent component analysis (ICA). Detecting Bad Epochs
The ICA is used to detect and correct artifacts, particular Hopefully, at this stage the EEG data is clean enough for further
eye movements, eye blinks and cardiac pulse artifacts (43). It is processing. Still, transient artifacts may remain (muscle artifacts,
important that the time course of the ICA components that are sweating, remaining eye blinks, etc.) that none of the steps above
considered to reflect one of these artifacts is inspected together successfully removed. It is therefore strongly recommended that
with the raw EEG data and it is assured that they indeed spatially the “cleaned” data are visually inspected and that bad epochs
(topography) and temporally correlate with the appearance of are marked. In Cartool, we have implemented a tool that helps
these events. Once this is assured, the data are back-projected to identify these bad epochs. It is based on a set of simple
by excluding these components. At the time of publication, ICA statistics on the tracks and then estimates how much each track
is not fully implemented in Cartool. An often used software for deviates from its own individual baseline. The statistics is based
artifact removal using ICA is EEGLAB (44). on instantaneous values (absolute value, variance, skewness and
After ICA correction, the bad electrodes detected in the first kurtosis among electrodes at a given time point) and on short
step are interpolated using a 3D or spherical spline algorithm time periods by computing the cross-convolution, which is a
(45). In order to do that, the 3-dimensional position of each convenient way to estimate the noise in a signal. All these outlier
electrode needs to be known (see section Determining the estimators are merged together to a single compound estimator
Solution Points in the Gray Matter.). and the highly suspicious time periods are highlighted. By visual
inspection, the user can then decide whether these periods should
be marked as “bad” or not. These bad epochs will be conveniently
Spatial Filtering used in later processing, as many toolboxes of Cartool allow to
The precursor of EEG source imaging is the scalp potential skip them.
map (5). Therefore, visualizing and inspecting the quality of the
topography of the maps is as important as the inspection of
the waveforms. Even after interpolation of artifacted electrodes CONSTRUCTING THE HEAD MODEL
and removing irrelevant ICA components, transient events can
The head model is the model for which the EEG forward solution
corrupt a few electrodes for a short time period. They can be seen
is calculated. The forward solution determines how much a
on the potential map displays as isolated “islands” within the local
given electrical source in the brain will impact each electrode
neighborhood. Such outlier electrodes will have dramatic effects
on the scalp. It provides the lead field matrix from which the
on source localization as the steep gradients will lead to local
inverse problem will be solved. It is strongly recommended
maxima beneath the electrode [see Figure 4.7. in (46)].
to use the individual MRI of the participant to construct the
Here we describe a spatial filter that we designed and
head model, particularly in clinical studies where the source
implemented in Cartool. It is an instantaneous filter which
localization is used to guide surgery as for example in epilepsy or
removes local outliers by spatially smoothing the maps without
in functional mapping of eloquent cortex. If this is not available,
losing its topographical characteristics.
a template MRI can be used (for example the MNI brain), but
The spatial filter is designed in the following
the source localization will be less precise, as shown in Brodbeck
way (see Figure 1A):
et al. (47) in a large patient cohort. The MRI needs several
• For each electrode, the values of the 6 closest neighbors are pre-processing steps in order to get to a proper delineation
determined, plus the central electrode value itself. of the gray matter in which the source activity is estimated,
• The 7 data points are sorted. and to describe the different compartments of the head (skin,
• The minimal and maximal values are removed by dropping the skull, CSF, brain) that have different conductivity parameters.
first and last items of this list. Since the electric field that spreads from the sources to the
• The remaining 5 values are then averaged, with weights scalp surface is attenuated by these compartments (particularly
proportional to the inverse distance to the central electrode. by the skull), a proper incorporation of the head shape and
The central electrode is given a weight of 1. the conductivity parameters in the head model is essential for

Frontiers in Neurology | www.frontiersin.org 4 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 1 | Illustration of the spatial filter implemented in Cartool. (A) Determination and removal of the maximal and minimal value of the 6 nearest neighbor of a
given electrode. (B) Illustration of the waveforms and the map (C) at a given time point before filtering. (D) Illustration of the effect of the spatial filter on the waveforms
and maps (E).

EEG source reconstruction. Once the MRI is pre-processed, the very detrimental for any further 3D processing, like filtering, and
electrodes have to be positioned on the head corresponding to needs to be addressed as a very first step. In Cartool this is done by
how they were positioned during the recordings. It is obvious simply up-sampling the lowest resolution axes with some linear
that if the position of the electrodes does not correspond to the rescaling, to end up with the highest resolution in all 3 axes. Once
real position from which the signal was recorded during the the MRI is made isotropic, the axes have to be re-oriented in a
experiment, the source localization will not be correct. standard way to improve readability and compatibility with other
software. As a default, Cartool transforms the MRI to the Right-
MRI Processing Anterior-Superior (“RAS”) orientation for the three axes X, Y, Z
The head model for EEG source imaging is based on the MRI. As (right-hand system) similar to the MNI template brain. This is
mentioned above, whenever possible the individual MRI should done by appropriate 90 degrees rotations (Figure 2B).
be used. It gives information about the shape of the head, the Once the main axes have been set, adjustments are performed
thickness of the skull and the volume of the gray matter within to further improve readability and comparison across subjects,
which the solution points for the source localization are defined. or comparison with the MNI template. First, an optimal sagittal
Several processing steps are needed in order to properly extract cutting plane is determined by adjusting 2 rotations values, on
this information. This includes re-sampling and re-orientation, the Y and Z axes, and 1 translation value on the X axis, until
skull stripping, Bias Field correction and separation of gray and the two halves defined by this plane are most symmetrical. This
white matter. These processing steps are fairly standard and is of utter importance for the later stage when laying out the
offered in many different software packages, most well-known in solution points in the brain, because it keeps an anatomically
the SPM toolbox (48). While Cartool allows to read MRI images realistic balance between the left and right hemispheres. Once the
and gray masks that have been processed by other software optimal sagittal plane has been found, the best transverse plane
programs, it also has an integrated MRI processing toolbox. It is determined. This is highly recommended as the placement
takes particular care of points that are crucial for a proper layout of the participant in the MR scanner varies. A tilted head is
of the solution points, such as assuring that no holes appear normalized in Cartool by adjusting 1 rotation value on the X
in the gray matter mask and that the sagittal plane is properly axis, and 2 translation values in Y and Z. The optimal transverse
determined to assure symmetry of the left and right hemisphere. adjustment is the one that gives a mid-sagittal plane that is
In the following, the way these processing steps are implemented most similar to the corresponding mid-sagittal plane of the MNI
in Cartool are described: head. This is done by tilting the head and setting the origin
above the anterior commissure. Note that these two steps partly
Re-sampling and Re-orientation solve the co-registration from a given head to the MNI template
Depending on how the MR scanner performed the acquisition (Figure 2C). Only a final rescaling (3 parameters) is needed
and how the participant lied in the scanner, re-sampling to achieve the ultimate 9 parameter co-registration. The last
and re-orientation of the MRI is needed as a first step. geometrical transform is to resample the MRI to reach the desired
In Cartool, the following geometrical transformations are all target voxel size, which is usually 1 mm3 .
built into a 4 × 4 affine transform matrix, which stacks All the steps above are then applied at once on the original
efficiently the successive steps described below, all of them being MRI, through a 4 × 4 affine transform matrix. Interpolation
mathematically linear. between the voxels is done with a Lanczos filter, with kernel
If the acquisition was anisotropic, as is often the case, voxel of size 3, which considers a neighborhood of 216 voxels for
sizes are not equal in all three dimensions (Figure 2A). This is each value to be interpolated. The target MRI size is optionally

Frontiers in Neurology | www.frontiersin.org 5 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 2 | Illustration of the MRI processing pipeline. (A) Original anisotropic MRI. (B) Result of up-sampling and re-orientation, with red, green, and blue axis
pointing, respectively to X, Y, Z. (C) Adjustment of the cutting planes and setting of the AC origin. (D) Result of the skull-stripping to isolate the brain. (E) Brain slices
which exhibit the Bias Field of the original MRI. (F) Same brain slices post-Bias Field Correction. (G) Extraction of the Gray matter.

optimized to include only the transformed head, plus some by the final histogram of the brain, which shows very clear-cut
margin, and drop any useless empty spaces. tissue separation.

Gray Matter Segmentation

Skull-Stripping and Bias Field Correction
The final step of the MRI processing is the separation of white
At this point, we should have a standardized individual head.
and gray matter. This is needed because EEG source localization
The next step in the pipeline is the skull-stripping to separate
usually restricts the source space to the gray matter that contains
skull, CSF and brain (Figure 2D). Two methods are available
the synapses where postsynaptic potentials can be generated.
in Cartool, one mainly based on morphological operators, the
Cartool extracts the gray mask by estimating the global intensity
other one on region growing. Both methods were designed for
distributions of the gray and white matter and the CSF with
T1 MRIs, but appear to be resilient enough to work on T2 or
a Mixture of Gaussians. It classifies each voxel by weighting
MP-RAGE images.
the Gaussian probabilities, based solely on the voxel intensity,
MRI scans usually have inhomogeneities in space, called Bias
with some neighborhood likelihood (for a given voxel, the
Field. Without correcting for it, a given brain tissue like the
greyer the neighbors, the higher the chance to be gray, too).
gray matter will have different values depending on its physical
Finally, morphological smoothing operators are applied to fill any
position in the scanner (Figure 2E). This is definitely a non-
possible holes in the gray mask. Note that the produced gray mask
desired property which will hamper the segmentation of the brain
is therefore slightly thicker than the actual gray matter, which can
into its constituent tissues. Cartool corrects for the Bias Field
be quite thin in some brain areas. The smoothing assures that no
of the segmented brain by iteratively equalizing the histogram
gray matter parts are missed (Figure 2G).
of the white matter across all 3D directions. Since the white
matter has the highest intensity values, it is a good marker for
inhomogeneities. Any variations across a given axis are attributed Determining the Solution Points in the
to the Bias Field and are corrected (Figure 2F). By repeating Gray Matter
this process across all directions, a global approximation of the The volume that has been obtained through the gray matter
Bias Field is determined. The validity of this method is reassured extraction is called the solution space, and constitutes the volume

Frontiers in Neurology | www.frontiersin.org 6 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

in which the electric activity will be allowed to be localized. missing part. In order to avoid such effects, Cartool smoothens
The solutions space will typically contain 3000–6000 individual the gray matter mask as described above to assure continuity
solution points and is thus basically a down-sampled version of of the solution points (Figure 3). This is a desired property
the gray mask. Because of the Nyquist theorem, down-sampling and not a defect.
should be done with some prior smoothing to prevent aliasing 4) Number of Solution Points
effects. Otherwise, this would result in missing solution points or The number of solution points is defined by the user,
discontinuities in areas where the spatial frequency is higher than with a recommended range between 3000 and 6000. There
the down-sampled spatial frequency. are obvious pros and cons for both low and high number of
In Cartool, finding the optimal solution point distribution is solution points (Table 2).
done in the following way, given a number of solution points
While computer speed is nowadays only a marginal problem,
to attain:
memory limitations can still be an issue. Numerical precision
1) An initial down-sampling factor is estimated. issues come from the fact that inverting large matrices will
2) The gray mask is down-sampled by the current factor, while cumulate more errors than smaller ones. The spatial resolution
remaining centered on the optimal center. This keeps the left (grid spacing) and accuracy (to be spot-on) is a sensitive problem.
and right parts as symmetrical as possible. More points mean more spatial resolution because of smaller grid
3) Solution points with <8 neighbors out of a neighborhood of spacing. This increases accuracy but only up to a limit. Accuracy
26 are removed, repeatedly for 3 times. will stop improving after a given number of solution points (i.e.,
4) The remaining solution points are counted. the inverse solution is not “getting better”) due to the fact that
5) If the count is close enough to the requested amount of the quantity of information that is put into the system remains
solution points, the process is stopped. Otherwise, the down- the same, and is set by the number of electrodes. Also, the matrix
sampling factor is up- or down- regulated according to the inversion process can intrinsically provide only a limited level
current numbers, and the process is repeated. of accuracy.
The solution point extraction is an important step of source
localization, very often overlooked, if not totally ignored in the NUMBER AND POSITIONING OF
literature. Here are some points that have to be considered: THE ELECTRODES
1) Left-right distribution Electrode Layout
As described above, the MRI has to be realigned to the What is the minimal number of electrodes needed for reliable
mid-sagittal plane. That means that the geometrical center source localization? This question is often asked, particularly
of the MRI is going through the YZ plane that cuts the from the clinical community that intends to apply EEG source
brain in two optimally symmetrical parts. When down- localization to the EEG that is routinely recorded with the
sampling the gray matter into the solution points, the new standard 10-20 system, i.e., with only 19 electrodes. Several
down-sampled center has to remain in this plane. This will studies have demonstrated that this low number not only leads
ensure that the resulting solution points will be equally to blurring of the solution, but also to incorrect localization (49)
distributed between the left and right hemisphere. Having compared the effective spatial resolution of different electrode
an asymmetrical distribution of solution points will have an montages (19-129 electrodes) and concluded that “the smallest
impact on the source localization by giving more weights to topographic feature that can be resolved accurately by a 32-
one side of the brain and attributing sources to the wrong side. channel array is 7 cm in diameter, or about the size of a lobe of
Obviously, a real asymmetrical (pathological) brain will have the brain”. Simulation studies as well as subsampling studies in
its mid-sagittal plane set according to its anatomy, and will epileptic patients with known epileptic focus clearly showed that
have an asymmetrical distribution of solution points. electrode arrays with <32 sensors lead to severe mislocalizations
2) Minimum neighborhood and blurring (3, 28). The significant increase in localization
The inverse process will later need the computation of a precision has been demonstrated by Brodbeck et al. (47) in a large
discrete Laplacian in the Solution Space. To be able to do that
correctly, each solution point has to have enough neighbors. In
Cartool, a quite conservative minimum of at least 8 neighbors TABLE 2 | Pros and Cons of the number of solution points in the inverse space.
out of 26 is chosen. Solution points that have less neighbors
Lower number of solution points Higher number of solution points
will be removed.
3) Continuity (+) Faster to compute the matrices (–) Longer to compute the matrices
The solution points should cover the whole gray matter (+) Less memory (–) More memory
without missing points on the thinner parts. It is obvious (+) Less numerical precision issues (–) More numerical precision issues
that source activity cannot be reconstructed on non-existent (+) Smaller matrices and faster display (–) Larger matrices and slower display
solution points, leading to a lack of precision for some brain (–) Less spatial resolution (+) More spatial resolution
areas. Another risk of missing solution points in some gray (–) Less spatial accuracy (+) Somewhat more spatial accuracy
matter parts is that neuronal activity coming from this area (–) Less neighbors around each solution (+) More neighbors around each
would be attributed to the solution points closest to the point solution point

Frontiers in Neurology | www.frontiersin.org 7 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 3 | Illustration of the distribution of the solution points in the gray matter.

group of epileptic patients where sensitivity and specificity were Ultimately, the position should correspond to the reality,
compared between high-density (128–256 channels) and routine i.e., to the actual position of the electrodes during the
clinical (19–21 channel) EEG. In a cohort of patients with focal recording, as this has a direct impact on the stability of the
ischemic stroke, (50) demonstrated that more than 64 electrodes source localization.
were needed to avoid mislocalizations of the affected regions. There are different levels of knowledge of the electrode
Recent studies showed that the detection and localization of high positions during the recordings. Nowadays, EEG caps or nets
frequency oscillations, which are potential markers of epileptic are usually used, with the advantage of fixed spacing between
areas, are better detected and localized with high- as compared electrodes. Many studies rely on these fixed positions determined
to low-density EEG (51, 52). Also, localization of seizure onset by the manufacturer and the names of the electrode according
zone using connectivity analysis in the source space was shown to the 10-10 coordinate system. A template 3D-array (often
to be more precise with high- compared to low-density EEG (53). provided by the manufacturer) is then used and it is assumed that
The fact that the skull resistance is much lower than previously the EEG cap is placed and adjusted according to some fixed points
assumed (see section The LSMAC Head Model), additionally (Inion, Nasion, preauricular points, Vertex, etc.). It is crucial
supported the notion that high-density EEG is needed to avoid that this placement is done correctly and it is recommended
spatial aliasing, that then leads to mislocalization (54, 55). As the that photographs are taken to later assure correspondence of the
skull is much thinner in babies, even more electrodes are needed electrodes to these fixed points when landing the electrode array
in this population (56, 57). on the MRI head. A more advanced and recommended method,
Nevertheless, these results do not necessarily mean that if available, is to measure the actual position of each electrode
imperfect spatial sampling precludes source localization. for each participant using a 3D digitizer or a photogrammetry
Even with <32 electrodes, source localization allows to gain system (65). The obviously most accurate method is to put
valuable insight about the underlying sources, particularly in the participant in the scanner with the cap on the head and
applications with well-defined focal activity such as epileptic afterwards mark the artifacts induced by the electrodes on the
spikes (15, 58–60). MR images (Figure 4A). This last method bypasses the co-
Besides the number of electrodes, their positioning in terms registration procedure described below. However, as it requires
of coverage of the head plays an important role too. The an MRI scanner close to the EEG recording room and MRI-
standard 10–20 system does not include electrodes over the compatible EEG caps, this method is rarely possible, except in
inferior part of the head which disfavors the proper recording simultaneous EEG-fMRI studies.
of activities in the inferior-basal and anterior part of the
temporal lobe where activity originating or propagating from
the mesial temporal structures is maximal (61, 62) (Figure 4B). Co-registration of the Electrodes on the
Missing these electrodes can lead to mislocalization of activities MRI Head
originating from the mesial temporal lobe (60, 63). It has In Cartool, the co-registration of the 3D electrode array is done
therefore been recommended that at least 3 inferior electrodes interactively by displaying and manually adjusting the global 3D
on each side should be added to the standard 10–20 system in shape of the electrode array to the shape of the head. This is
clinical routine (64). a way to make use of all the available geometrical information,
instead of relying only on a few fiducial positions. The method
3D Electrode Positions can adapt to all cases and allows to co-register either an individual
The correct positioning of the electrodes on the surface of or a template electrode array to either an individual or a template
the head of the participant’s MRI is an important point. MRI head.

Frontiers in Neurology | www.frontiersin.org 8 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 4 | (A) Example of the location of 256 electrodes on the head determined by the artifacts that the electrodes create on the MRI image by wearing the EEG
net in the scanner. (B) Location of the electrodes with respect to the brain: Blue: 256 electrode net. Red: Positions of the 19 electrodes of the standard clinical 10–20
system. The zoomed-in regions show the bad coverage of the frontal, basal temporal and midline areas with the 19 electrodes as compared to the 256 electrodes.

In detail, the following steps are performed interactively: CALCULATING THE LEAD FIELD
Both the processed (resampled and reoriented) MRI and the
electrode array are displayed on the screen. The operator In order to calculate the lead field, a head model has to be created
then virtually adjusts the electrode array on the MRI head, that incorporates as realistically as possible the shape of the head
mimicking the way the physical electrodes were set on and the conductivity parameters of the different tissues between
the subject’s head. This is done by shifting the electrode the current sources in the brain and the potential on the scalp.
positions in any direction, rotating and stretching them There have been substantial advancements in the construction
until they convincingly look like the reality. Photographs of realistic head models. Still, even the most sophisticated
taken during the recording can help to properly adjust methods are simplified descriptions of the complex organization
the positions. of head tissues. The often-used realistic models are the Boundary
Once this adjustment is done, Cartool provides a last useful Element Model (BEM) and the Finite Element Model (FEM).
feature: virtually “gluing” the electrodes on the head. For many Their superiority compared to 3-shell spherical head models has
reasons, like a template electrode array being used on a real MRI been demonstrated in simulations (66–68) as well as real data
head, and no matter how much care is devoted to the previous (69, 70). The downside of these sophisticated head models is
steps, many electrodes can end up either being below or above the an increased computational load because numerical solutions
scalp surface. This in turns will be detrimental to the Lead Field have to be applied. They are also more sensitive to any mishap
computation by biasing the distances from any given electrode happening during the brain and gray matter extraction, as more
to the brain. By activating this virtual gluing, all electrodes will brain tissues and more parameters are involved. In Cartool, we
be perfectly projected perpendicularly on the nearest position on implemented a method that we called Locally Spherical Model
the scalp (Figure 5). with Anatomical Constraints (LSMAC, see below). It tries to

Frontiers in Neurology | www.frontiersin.org 9 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 5 | Original position of a template electrode layout with respect to the head of the subject (left) and the corrected positions after manual rotation and
translation and the final automatic “gluing” on the scalp.

counteract the computational cost of the BEM and FEM models

by using analytical equations while still keeping the realistic
aspect of the head geometry and the local variability of the
thickness of the skull. Birot et al. (71) compared the LSMAC
model with a BEM and a FEM model in a dataset of 38
epileptic patients in whom high-density scalp EEG, intracranial
EEG and localization of the resection brain area that rendered
the patient seizure-free was available. LSMAC, BEM and FEM
were computed from the individual MRI of the patients and
source localization was performed on averaged interictal epileptic
discharges. Similar source location accuracy with respect to the
intracranial recordings and the resected zone was found for
all three head models. It was concluded that in such clinical
applications, the use of highly sophisticated and difficult to
implement head models is not a crucial factor for accurate
source localization.

The LSMAC Head Model

The Locally Spherical Model with Anatomical Constraints
(LSMAC) (29) is an adaptation of the SMAC head model
introduced by Spinelli et al. (24). The LSMAC Lead Field
calculation requires the pre-processed full head and the gray
mask MRIs, the co-registered electrodes and the location of FIGURE 6 | Illustration of the determination of the skull thickness under each
electrode. A sagittal cutting plane is shown with the electrodes (in blue) located
the solution points. Under each electrode, the inner and outer
on the scalp surface, the radius lines (in yellow) extending from the center to
borders of the skull are then automatically determined and the each electrode, and on each line three dots showing where the skull and scalp
global resistivity value is locally corrected. This decreases the limits are determined.
sources of error in EEG inverse modeling. The borders of the
skull are determined by analyzing the gray levels of a radial line,
going from the center of the brain to the electrode on the scalp. variability is adjusted. Figure 6 shows an example of the skull
Since the skull is barely visible in T1 MRI scans, it shows up as radii estimation on 3 electrodes.
dark voxels in contrast to the scalp and the brain. Consequently, These skull radii estimate still has some uncertainty due to
the beginning and end of the skull can be identified as borders the nature of the MRI T1 images. To further increase their
between light and dark voxels on the line. By measuring these precision, Cartool requests the user to provide a target age of
borders repeatedly with slight offsets on the scalp, uncertainty the subject. Using thickness values described in the literature for
pertaining to noise, low voxel intensities and bone structure different age ranges (72, 73) and linear interpolation of missing

Frontiers in Neurology | www.frontiersin.org 10 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

values, a curve of the estimated mean thickness for each age electrocortical stimulation, fMRI, and clinical outcome after
was built (Figure 7A). The radii determined from the MRI are surgery [e.g., (47, 80–84)].
then globally rescaled to reach the estimated mean thickness for
the given age. This adjustment allows a better estimation of the Regularization Optimization
Lead Field in children, or in the difficult case of newborns. A Tikhonov regularization is typically used in the case of under-
second advantage of this adjustment is to allow the computation determined system of equations, such as when inverting the
of a Lead Field for any specific age from a fixed template, if the Lead Field. Simply put, it factors in the equations a level of
individual MRI was not available. EEG noise, and enforces a level of smoothness in the inverted
The skull resistivity has been shown to be much lower than results. The more regularization, the smoother the results and the
previous literature suggested. The resistivity ratio between the less the sensitivity to noise. However, too much regularization,
brain and the skull is around 1:10 to 1:30 (74–76), rather by over-smoothing the results, will degrade the accuracy of
than 1:80, as previously assumed (77). Also, the skull resistivity the localization. We wish to use the most precise amount of
increases with age. Cartool thus has a second built-in curve regularization despite the fact that the amount of noise is
that gives the relative conductivity of the skull compared to not known in advance, and will vary from case to case. To
the adjacent tissue (brain, CSF, scalp) as a function of age. handle all cases, Cartool computes 13 matrices with increasing
The curve is based on a few reported resistivity measures regularization factors from 0 (none) to 12 (for very noisy data)
of living tissue (78, 79). According to these reports, the times a constant ∝, which depends on the selected inverse model.
conductivity ratio varied between 1:9.80 (11 years old) to The stack of 13 matrices is then saved into a single file. Later on,
1:25 (50 years old). An additional estimated value of 1:50 when applying an actual EEG to the inverse matrix, its noise level
for 100 years old was added to be able to extrapolate the will be evaluated, and the optimal matrix will be chosen.
curve past 50 years old, basically following the decreasing The general equation for the inverse problem with Tikhonov
trend. When the age of the subject is entered in Cartool, the regularization can be written as:
conductivity value is adapted to this age according to the curve
+
shown in Figure 7B. J = W.K t .(K.W.K t + ∝R .I) .8 (2)

CALCULATING THE INVERSE SOLUTION With J being the source density, 8 the electric field, K the
Lead Field, W some specific inverse weighting factors and I the
The inverse problem has no unique solution and a priori identity matrix.
assumptions have to be incorporated to derive to a unique The regularization factor ∝R is set the following way, for R
assumption of the distribution of neuronal activity in the brain varying from 0 to 12:
that lead to a certain potential field on the scalp. As explained in
the Introduction, a number of solutions of the inverse problem ∝R = R. ∝
max ( Eigenvalues K.W.K t )


have been proposed, incorporating different constraints based
on a priori information about the desired source characteristics ∝ = (3)
20000
or on physiological assumptions [for comprehensive reviews see
(3, 25–28)]. In Cartool, we implemented three linear distributed The optimal regularization for a given EEG is defined as the L-
source models: the weighted minimum norm solution (21) the corner of the norm of the solution points as a function of the
low resolution electromagnetic tomography [LORETA; (22)], regularization factor R. When Cartool applies the inverse matrix
and the Local AUtoRegressive Average [LAURA; (23)], all being to the data, it automatically defines this L-corner over the whole
modifications of the minimum norm (NM) solution (18). We dataset and uses this optimal regularization factor for all time
validated these implementations in several experimental and points. Alternatively, the user can specify a certain regularization
clinical studies by comparing them with intracranial recordings, factor for each dataset.

FIGURE 7 | Age correction of skull thickness and skull conductivity. (A) Estimated average skull thickness across age. (B) Estimated skull conductivity ratios across
age.

Frontiers in Neurology | www.frontiersin.org 11 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 8 | Source localization normalization. (A) Time series of 3 solution points showing the difference in mean amplitude (norm) between them. (B) The same time
series as in (A) but after normalization, showing that the 3 solution points now have the same amplitude range. (C) Histograms of these 3 solution points, showing that
the background activity is the left-most mode of the distribution. (D) Histogram after normalization, showing that all the background activity has been centered to 1.
(E) Histograms for all solution points (vertical axis), with the red color coding for the highest source amplitude probability. (F) Histogram of source amplitude probability
for each solution point after normalization, showing that all solution points now have a background range from 0 to 1, while retaining their respective highest activities.

Normalization of the Inverse for example in averaged epileptic spikes or in evoked potentials
Solution Result restricted to the time period of sensory processing or motor
When inspecting the estimated current density at each solution responses. The normalization should therefore be applied to non-
point across time in ongoing (non-averaged) EEG it appears that averaged raw data transformed to the source space. A recent
substantial variability of power is observed across solution points. study where this normalization method has been used on resting-
These variations are supposed to come from geometrical and state EEG to determine the sources of the EEG microstates in
mathematical approximations that are done during the different task-induced, self-initiated thoughts, showed that this method
steps of the inverse matrix calculation. It is thus necessary to reveals brain networks that overlap with those derived by fMRI
find a way to correct for this power variability, in order to in the same subjects (85).
reliably estimate the fluctuations of brain activity over time in Here is a step-by-step description of this
individual subjects and to compare them between subjects. In specialized normalization:

Cartool, we implemented a normalization approach by using the Given a 3D dipole spx , spy , spz at a given solution point sp,
background activity of the norm of the inverse solution over time we define spχ as the squared value of its norm:
to estimate a baseline and a scaling factor for each solution point.
In order to have a robust estimation, a large enough time sample
should be used, preferably the whole pre-processed and artifact- spχ = spx 2 + spy 2 + spz 2 (4)
excluded data of a given subject. Still, the correction factors can
be satisfactorily computed on as little as a thousand time points,
as long as no solution point remains in the same stable state The noisy part of the data therefore follows a Chi-square
more than half of the sampled time, which might be problematic distribution of degree 3.

Frontiers in Neurology | www.frontiersin.org 12 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 9 | Illustration of the actual vectorial results (top left, in 3D) of the distributed sources, and their corresponding amplitude values (top right, in 3D, and bottom
as transverse slices).

The variable spχ can be approximated to a normally Both the Mode and the MAD being computed on the left part of
distributed variable spN by (86): the probability density function is key here. In this way it ignores
any activity above noise level that might be present in some brain
spN = spχ 0.2887 (5) areas while not in others. Rescaling using the actual activities
would be incorrect, as it would basically transform them into the
Having now a normal distribution, spN can be standardized into baseline. On the other hand, noise can be seen on all solution
spZ by using the regular z-transform: points and its level is a good estimator of the rescaling that has to

be applied. Implementation-wise, these estimators are computed
spN − µspN multiple times on random sub-samplings of the data, and the two
spZ = (6)
σspN respective medians of all these estimators are finally taken.
Finally, because we started with positive data (the norm of a
However, the values of µ and σ used for the z-transform dipole), we also wish to end up with positive data in order to avoid
have to be calculated only on the noisy part of the data—the any confusion due to having signed results. We define spZ+ as spZ
background activity from the Chi-square i.e., the lowest values of shifted by 3 standard deviations to the right, then divided by 3 so
the probability density function. Hence µ is estimated from the that the background mode is finally aligned to 1.
left-most Mode of the spN distribution:

\
µspN = Mode left (spN ) (7)

spZ+ = max( spZ + 3 /3, 0) (9)
For the same reason, σ is estimated from the Median of Absolute
Deviation (MAD), centered on the previously estimated µ, and
computed only with the values below µ: After this standardization procedure, the power of the current
density is comparable across all solution points, and its noisy
\
σspN = MAD left (spN ) (8) component is normally distributed (Figure 8).

Frontiers in Neurology | www.frontiersin.org 13 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

FIGURE 10 | Illustration of the visualization of the data and the results of the different analysis steps as implemented in Cartool. All windows can be independently
manipulated in 3D. The screen shot shows a visual evoked potential (face presentation) recorded with 256 electrodes, the corresponding potential map at 188 ms
post-stimulus and the estimated sources located in the mesial temporal lobes and the fusiform gyrus.

Results of Inverse Solution: Vectorial added value of this method in the pre-surgical assessment of these
vs. Scalar patients has been demonstrated repeatedly, not only for focus
The output of the inverse matrix multiplication with the EEG localization, but also for localization of eloquent cortex (3, 47, 83,
results in equivalent dipoles located on each solution point. As 91–93). Besides the clinical significance, EEG source localization
each dipole is a 3D vector, it is described as an amplitude in in epilepsy also gives the unique possibility to evaluate the
the x-,y-, and z- directions. In most applications, this vectorial performance and precision of different head- and source-models
information is not relevant and only the norm (amplitude) of the because intracranial recordings or the outcome after surgery
dipoles is saved, i.e. scalar values. This results in positive values at can serve as “gold-standard” (71, 82, 94, 95). The most direct
each solution point as displayed in Figure 9. way to evaluate EEG source localization is the simultaneous
recording of scalp- and intracranial EEG. A recent study with
high-density (256-channel) scalp EEG recorded simultaneously
APPLICATIONS OF EEG with intracranial local field potentials from deep brain structures
SOURCE LOCALIZATION in patients undergoing deep brain stimulation demonstrated that
EEG source localization is able to sense and properly localize
High-density EEG recordings have become standard in many spontaneous Alpha activity generated in the thalamus or the
experimental as well as clinical laboratories, given that most nucleus accumbens (84). This demonstration opens new doors
manufactures readily provide such systems and that the in the use of high-density EEG source imaging, as it shows that
application of many electrodes has become fairly easy. It source localization is not restricted to the cortex only.
therefore does not come as a surprise that EEG source localization In experimental studies, EEG source imaging has become
is increasingly used to infer to the areas in the brain that standard to localize different brain areas involved in sensory,
generated the activity observed on the scalp (87–89). Concerning motor, and cognitive functions, most often applied to event-
clinical applications, the undoubtedly most intense use of EEG related potentials (89, 96). However, EEG source imaging is
source localization is in epilepsy, with the intention to localize the also increasingly used to define large-scale network dynamics
epileptic zone in pharmaco-resistant focal epilepsies (3, 90). The by applying connectivity measures (97–99). Because of the high

Frontiers in Neurology | www.frontiersin.org 14 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

temporal resolution of EEG, functional connectivity measures reference for those who use Cartool. In view of the increasing
such as Granger Causality methods are used to study directional practice of source localization in EEG and MEG applications, it is
connectivity of large-scale networks in the healthy (100–102) important that the user well-understands how the software that
and in the pathological (103, 104) brain. It has thereby become he/she is using implement the different steps. We also consider
clear that such connectivity measures have to be applied in it of crucial importance that the data and the results of the
source space and not on the level of the scalp electrodes, analysis are visualized and that the user inspects the data carefully
since volume conduction and reference-dependency make the in all different steps and assures that the results make sense
interpretability of sensor-based connectivity measures difficult (Figure 10).
(105–110). Therefore, EEG source imaging is a pre-requisite This review also intends to make the user aware of the
for functional connectivity analysis [for a recent tutorial paper obstacles and limitations of each step of the analysis and the
on EEG connectivity measures see (111)]. It is thus of utmost choices that have to be made. Basic knowledge of the underlying
importance that the source localization is done properly and reasons for these choices and how it is implemented in a given
that the steps described in this review paper are understood and software is mandatory to avoid misinterpretation of the results
correctly applied. and to properly describe the methods in a publication. Finally,
we hope that this review contributes to the global awareness
CONCLUSIONS that EEG source imaging is feasible and doable even for non-
engineers and provides information about the function of the
This review describes in detail the different steps that are needed human brain that cannot be achieved by analysis restricted to the
to derive from a multichannel scalp EEG recording to the scalp level.
estimation of the distribution of the underlying neuronal sources.
It explains the logic underlying each step and the requirements AUTHOR CONTRIBUTIONS
that need to be fulfilled to perform them. It illustrates how these
steps are implemented in one particular stand-alone software: All authors listed have made a substantial, direct and intellectual
Cartool. While this might occasionally give the impression of a contribution to the work, and approved it for publication.
software manual rather than a review paper, we do not intend to
claim that this software is the only one that allows to perform FUNDING
these steps adequately. Several other stand-alone or open-source
software packages exist, commercially or freely available, that This work is supported by the Swiss National Science Foundation
have implemented these analysis steps in similar or slightly (No. 320030_159705) and the Swiss National Center of
different ways (30); Table 1. We here use the example of Cartool Competence in Research; Synapsy: the Synaptic Basis of Mental
to illustrate the implementation and the usage and to provide a Diseases (NCCR Synapsy Grant # 51NF40-185897).

REFERENCES 9. Scherg M, von Cramon D. A new interpretation of the generators of BAEP

waves I-V: results of a spatio-temporal dipole model. Electroencephalogr Clin
1. Helmholtz HLP. Ueber einige gesetze der vertheilung elektrischer ströme in Neurophysiol. (1985) 62:290–9. doi: 10.1016/0168-5597(85)90006-1
körperlichen leitern mit anwendung aud die thierisch-elektrischen versuche. 10. Scherg M, Picton TW. Separation and identification of event-related
Ann Physik und Chemie. (1853) 9:211–33. potential components by brain electric source analysis. In: Brunia CHM,
2. Malmivuo J, Plonsey R. Bioelectromagnetism: Principles and Applications Mulder G, Verbaten MN, editors. Event-Related Brain Research. Amsterdam:
of Bioelectric and Biomagnetic Fields. New York, NY: Oxford University Elsevier (1991). p. 24–37.
Press (1995). 11. Mosher JC, Lewis PS, Leahy RM. Multiple dipole modeling and localization
3. Michel CM, He B. EEG mapping and source imaging. In: Schomer DL, Silva, from spatio-temporal MEG data. IEEE Trans Biomed Eng. (1992) 39:541–57.
FHLd, editors. Niedermeyer’s Electroencephalography. New York, NY: Oxford doi: 10.1109/10.141192
University Press (2018). p. 1135–156. 12. Henderson CJ, Butler SR, Glass A. The localization of equivalent dipoles of
4. Nunez PL, Srinivasan R. Electric Fields of the Brain: The Neurophysics of EEG. EEG sources by the application of electrical field theory. Electroencephalogr
2nd ed. New York, NY: Oxford University Press (2006). Clin Neurophysiol. (1975) 39:117–30. doi: 10.1016/0013-4694(75)90002-4
5. Fender D. Source localization of brain electrical activity. In: Remond 13. Ebersole JS. Non-invasive localization of the epileptogenic focus by
ASGA, editor. Methods of Analysis of Brain Electrical and Magnetic Signals. EEG dipole modeling. Acta Neurol Scan Suppl. (1994) 152:20–8.
Amsterdam: Elsevier Science Publishers (1987). doi: 10.1111/j.1600-0404.1994.tb05179.x
6. Nunez PL, Nunez MD, Srinivasan R. Multi-scale neural sources of EEG: 14. Lantz G, Holub M, Ryding E, Rosen I. Simultaneous intracranial and
genuine, equivalent, and representative. A tutorial review. Brain Topogr. extracranial recording of interictal epileptiform activity in patients with
(2019) 32:193–214. doi: 10.1101/391318 drug resistant partial epilepsy: patterns of conduction and results from
7. Kavanagh RN, Darcey TM, Lehmann D, Fender DH. Evaluation of methods dipole reconstructions. Electroencephalogr Clin Neurophysiol. (1996) 99:69–
for three-dimensional localization of electrical sources in the human 78. doi: 10.1016/0921-884X(96)95686-6
brain. IEEE Trans Biomed Eng. (1978) 25:421–9. doi: 10.1109/TBME.1978. 15. Sharma P, Scherg M, Pinborg LH, Fabricius M, Rubboli G, Pedersen B,
326339 et al. Ictal and interictal electric source imaging in pre-surgical evaluation:
8. He B, Musha T, Okamoto Y, Homma S, Nakajima Y, Sato T. Electric a prospective study. Eur J Neurol. (2018) 25:1154–60. doi: 10.1111/ene.
dipole tracing in the brain by means of the boundary element 13676
method and its accuracy. IEEE Trans Biomed Eng. (1987) 34:406–14. 16. Willemse RB, Hillebrand A, Ronner HE, Vandertop WP, Stam CJ.
doi: 10.1109/TBME.1987.326056 Magnetoencephalographic study of hand and foot sensorimotor

Frontiers in Neurology | www.frontiersin.org 15 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

organization in 325 consecutive patients evaluated for tumor or epilepsy 37. Grooms JK, Thompson GJ, Pan WJ, Billings J, Schumacher EH, Epstein CM,
surgery. Neuroimage Clin. (2016) 10:46–53. doi: 10.1016/j.nicl.2015.11.002 et al. Infraslow electroencephalographic and dynamic resting state network
17. Stefan H, Hummel C, Scheler G, Genow A, Druschky K, Tilz C, et al. activity. Brain Connect. (2017) 7:265–80. doi: 10.1089/brain.2017.0492
Magnetic brain source imaging of focal epileptic activity: a synopsis of 455 38. Frauscher B, von Ellenrieder N, Zelmann R, Rogers C, Nguyen DK, Kahane
cases. Brain. (2003) 126:2396–405. doi: 10.1093/brain/awg239 P, et al. High-frequency oscillations in the normal human brain. Ann Neurol.
18. Hämäläinen M, Ilmoniemi RJ. Interpreting measured magnetic fields of the (2018) 84:374–85. doi: 10.1002/ana.25304
brain: minimum norm estimates. Med Biol Eng Comput. (1994) 32:25–42. 39. Rousselet GA. Does filtering preclude us from studying ERP time-courses?
19. Grave de Peralta Menendez R, Gonzalez Andino SL. A critical analysis Front Psychol. (2012) 3:131. doi: 10.3389/fpsyg.2012.00131
of linear inverse solutions. IEEE Trans Biomed Eng. (1998) 45, 440–448. 40. Widmann A, Schroger E. Filter effects and filter artifacts in the
doi: 10.1109/10.664200 analysis of electrophysiological data. Front Psychol. (2012) 3:233.
20. Greenblatt RE. Probabilistic reconstruction of multiple sources in the doi: 10.3389/fpsyg.2012.00233
neuroelectromagnetic inverse problem. Inverse Problems. (1993) 9:271–84. 41. Smith SW. The Scientist and Engineer’s Guide to Digital Signal Processing. 2nd
doi: 10.1088/0266-5611/9/2/008 ed. San Diego, CA: California Technical Publishing (1999).
21. Wang JZ, Williamson SJ, Kaufman L. Magnetic source images determined 42. Hogenauer E. A class of digital filters for decimation and interpolation. In:
by a lead-field analysis: the unique minimum-norm least-squares ICASSP ’80. IEEE International Conference on Acoustics, Speech, and Signal
estimation. IEEE Trans Biomed Eng. (1992) 39:665–75. doi: 10.1109/ Processing. (1980). p. 271–4. doi: 10.1109/ICASSP.1980.1170846
10.142641 43. Delorme A, Sejnowski T, Makeig S. Enhanced detection of artifacts in EEG
22. Pascual-Marqui RD, Michel CM, Lehmann D. Low resolution data using higher-order statistics and independent component analysis.
electromagnetic tomography: a new method for localizing electrical NeuroImage. (2007) 34:1443–9. doi: 10.1016/j.neuroimage.2006.11.004
activity in the brain. Int J Psychophysiol. (1994) 18:49–65. 44. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of
doi: 10.1016/0167-8760(84)90014-X single-trial EEG dynamics including independent component analysis. J
23. Grave de Peralta Menendez R, Murray MM, Michel CM, Neurosci Methods. (2004) 134:9–21. doi: 10.1016/j.jneumeth.2003.10.009
Martuzzi R, Gonzalez Andino SL. Electrical neuroimaging based 45. Perrin F, Pernier J, Bertrand O, Echallier JF. Spherical splines for
on biophysical constraints. NeuroImage. (2004) 21, 527–539. scalp potential and current density mapping. Electroencephalogr Clin
doi: 10.1016/j.neuroimage.2003.09.051 Neurophysiol. (1989) 72:184–9. doi: 10.1016/0013-4694(89)90180-6
24. Spinelli L, Andino SG, Lantz G, Seeck M, Michel CM. Electromagnetic 46. Michel CM, Brandeis D. Data acquisition and pre-processing standards for
inverse solutions in anatomically constrained spherical head electrical neuroimaging. In: Michel CM, Koenig T, Brandeis D, Gianotti LRR,
models. Brain Topogr. (2000) 13:115–25. doi: 10.1023/A:10266071 Wackermann J, editors. Electrical Neuroimaging. Cambridge: Cambridge
18642 University Press (2009). p. 79–92.
25. Baillet S, Mosher JC, Leahy RM. Electromagnetic brain mapping. IEEE Signal 47. Brodbeck V, Spinelli L, Lascano AM, Wissmeier M, Vargas MI,
Process. Magaz. (2001) 14–30. doi: 10.1109/79.962275 Vulliemoz S, et al. Electroencephalographic source imaging: a prospective
26. He B, Lian J. Electrophysiological Neuroimaging: solving the EEG inverse study of 152 operated epileptic patients. Brain. (2011) 134:2887–97.
problem. In: He B, editor. Neuroal Engineering. Norwell, MA: Kluwer doi: 10.1093/brain/awr243
Academic Publishers (2005). p. 221–61. doi: 10.1007/0-306-48610-5_7 48. Friston K, Ashburner J, Kiebel S, Nichols TE, Penny W.
27. He B, Sohrabpour A, Brown E, Liu Z. Electrophysiological source imaging: Statistical Parametric Mapping. Amsterdam: Elsevier Ltd. (2007).
a noninvasive window to brain dynamics. Annu Rev Biomed Eng. (2018) doi: 10.1016/B978-012372560-8/50002-4
20:171–96. doi: 10.1146/annurev-bioeng-062117-120853 49. Srinivasan R, Tucker DM, Murias M. Estimating the spatial Nyquist
28. Michel CM, Murray MM, Lantz G, Gonzalez S, Spinelli L, Grave de of the human EEG. Behav. Res. Methods Ins. Comp. (1998) 30:8–19.
Peralta R. EEG source imaging. Clin Neurophysiol. (2004) 115:2195–222. doi: 10.3758/BF03209412
doi: 10.1016/j.clinph.2004.06.001 50. Luu P, Tucker DM, Englander R, Lockfeld A, Lutsep H, Oken B.
29. Brunet D, Murray MM, Michel CM. Spatiotemporal analysis of Localizing acute stroke-related EEG changes: assessing the effects
multichannel EEG: CARTOOL. Comput Intell Neurosci. (2011) 2011:813870. of spatial undersampling. J Clin Neurophysiol. (2001) 18:302–17.
doi: 10.1155/2011/813870 doi: 10.1097/00004691-200107000-00002
30. Baillet S, Friston K, Oostenveld R. Academic software applications 51. Kuhnke N, Schwind J, Dumpelmann M, Mader M, Schulze-Bonhage
for electromagnetic brain mapping using MEG and EEG. A, Jacobs J. High Frequency oscillations in the ripple band (80-
Comput Intell Neurosci. (2011) 2011:972050. doi: 10.1155/2011/9 250 Hz) in Scalp EEG: higher density of electrodes allows for better
72050 localization of the seizure onset zone. Brain Topogr. (2018) 31:1059–72.
31. Tadel F, Baillet S, Mosher JC, Pantazis D, Leahy RM. Brainstorm: a user- doi: 10.1007/s10548-018-0658-3
friendly application for MEG/EEG analysis. Comput Intell Neurosci. (2011) 52. Zelmann R, Lina JM, Schulze-Bonhage A, Gotman J, Jacobs J. Scalp EEG is
2011:879716. doi: 10.1155/2011/879716 not a blur: it can see high frequency oscillations although their generators are
32. Delorme A, Mullen T, Kothe C, Akalin Acar Z, Bigdely-Shamlo N, small. Brain Topogr. (2014) 27:683–704. doi: 10.1007/s10548-013-0321-y
Vankov A, et al. EEGLAB, SIFT, NFT, BCILAB, and ERICA: new tools 53. Staljanssens W, Strobbe G, Holen RV, Birot G, Gschwind M, Seeck
for advanced EEG processing. Comput Intell Neurosci. (2011) 2011:130714. M, et al. Seizure onset zone localization from ictal high-density
doi: 10.1155/2011/130714 EEG in refractory focal epilepsy. Brain Topogr. (2017) 30:257–71.
33. Oostenveld R, Fries P, Maris E, Schoffelen JM. FieldTrip: Open doi: 10.1007/s10548-016-0537-8
source software for advanced analysis of MEG, EEG, and invasive 54. Malmivuo JA, Suihko VE. Effect of skull resistivity on the spatial
electrophysiological data. Comput Intell Neurosci. (2011) 2011:156869. resolutions of EEG and MEG. IEEE Trans Biomed Eng. (2004) 51:1276–80.
doi: 10.1155/2011/156869 doi: 10.1109/TBME.2004.827255
34. Dalal SS, Zumer JM, Guggisberg AG, Trumpis M, Wong DD, Sekihara 55. Ryynanen OR, Hyttinen JA, Malmivuo JA. Effect of measurement noise and
K, et al. MEG/EEG source reconstruction, statistical evaluation, and electrode density on the spatial resolution of cortical potential distribution
visualization with NUTMEG. Comput Intell Neurosci. (2011) 2011:758973. with different resistivity values for the skull. IEEE Trans Biomed Eng. (2006)
doi: 10.1155/2011/758973 53:1851–8. doi: 10.1109/TBME.2006.873744
35. Litvak V, Mattout J, Kiebel S, Phillips C, Henson R, Kilner J, et al. EEG and 56. Fifer WP, Grieve PG, Grose-Fifer J, Isler JR, Byrd D. High-density
MEG data analysis in SPM8. Comput Intell Neurosci. (2011) 2011:852961. electroencephalogram monitoring in the neonate (2006). Clin Perinatol. 33,
doi: 10.1155/2011/852961 679-691. doi: 10.1016/j.clp.2006.06.011
36. Pernet CR, Garrido M, Gramfort A, Maurits N, Michel C, Pang E. et al. Best 57. Grieve PG, Emerson RG, Isler JR, Stark RI. Quantitative analysis of spatial
practices in data analysis and sharing in neuroimaging using MEEG. (2018). sampling error in the infant and adult electroencephalogram. NeuroImage.
doi: 10.31219/osf.io/a8dhx (2004) 21:1260–74. doi: 10.1016/j.neuroimage.2003.11.028

Frontiers in Neurology | www.frontiersin.org 16 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

58. Baroumand AG, van Mierlo P, Strobbe G, Pinborg LH, Fabricius M, temporarily removed during epilepsy surgery. Brain Topogr. (2003) 16:29–
Rubboli G, et al. Automated EEG source imaging: a retrospective, 38. doi: 10.1023/A:1025606415858
blinded clinical validation study. Clin Neurophysiol. (2018) 129:2403–10. 79. Latikka J, Kuurne T, Eskola H. Conductivity of living intracranial tissues.
doi: 10.1016/j.clinph.2018.09.015 Phys Med Biol. (2001) 46:1611–6. doi: 10.1088/0031-9155/46/6/302
59. Ding L, Worrell GA, Lagerlund TD, He B. Ictal source analysis: localization 80. Brodbeck V, Lascano AM, Spinelli L, Seeck M, Michel CM. Accuracy of EEG
and imaging of causal interactions in humans. NeuroImage. (2007) 34:575– source imaging of epileptic spikes in patients with large brain lesions. Clin
86. doi: 10.1016/j.neuroimage.2006.09.042 Neurophysiol. (2009) 120:679–85. doi: 10.1016/j.clinph.2009.01.011
60. Sperli F, Spinelli L, Seeck M, Kurian M, Michel CM, Lantz G. EEG source 81. Brodbeck V, Spinelli L, Lascano AM, Pollo C, Schaller K, Vargas
imaging in pediatric epilepsy surgery: a new perspective in presurgical MI, et al. Electrical source imaging for presurgical focus localization
workup. Epilepsia. (2006) 47:981–90. doi: 10.1111/j.1528-1167.2006.00550.x in epilepsy patients with normal MRI. Epilepsia. (2010) 51:583–91.
61. Koessler L, Cecchin T, Colnat-Coulbois S, Vignal JP, Jonas J, Vespignani doi: 10.1111/j.1528-1167.2010.02521.x
H, et al. Catching the invisible: mesial temporal source contribution to 82. Megevand P, Spinelli L, Genetti M, Brodbeck V, Momjian S, Schaller
simultaneous EEG and SEEG recordings. Brain Topogr. (2015) 28:5–20. K, et al. Electric source imaging of interictal activity accurately localises
doi: 10.1007/s10548-014-0417-z the seizure onset zone. J Neurol Neurosurg Psychiatry. (2014) 85:38–43.
62. Rosenzweig I, Fogarasi A, Johnsen B, Alving J, Fabricius ME, Scherg doi: 10.1136/jnnp-2013-305515
M, et al. Beyond the double banana: improved recognition of temporal 83. Lascano AM, Grouiller F, Genetti M, Spinelli L, Seeck M, Schaller K,
lobe seizures in long-term EEG. J Clin Neurophysiol. (2014) 31:1–9. et al. Surgically relevant localization of the central sulcus with high-
doi: 10.1097/WNP.0000000000000019 density somatosensory-evoked potentials compared with functional
63. Bach Justesen A, Eskelund Johansen AB, Martinussen NI, Wasserman magnetic resonance imaging. Neurosurgery. (2014) 74:517–26.
D, Terney D, Meritam P, et al. Added clinical value of the inferior doi: 10.1227/NEU.0000000000000298
temporal EEG electrode chain. Clin Neurophysiol. (2018) 129:291–5. 84. Seeber M, Cantonas LM, Hoevels M, Sesia T, Visser-Vandewalle V,
doi: 10.1016/j.clinph.2017.09.113 Michel CM. Subcortical electrophysiological activity is detectable
64. Seeck M, Koessler L, Bast T, Leijten F, Michel C, Baumgartner C, et al. The with high-density EEG source imaging. Nat Commun. (2019) 10:753.
standardized EEG electrode array of the IFCN. Clin Neurophysiol. (2017) doi: 10.1038/s41467-019-08725-w
128:2070–7. doi: 10.1016/j.clinph.2017.06.254 85. Bréchet L, Brunet D, Birot G, Gruetter R, Michel CM, Jorge J. Capturing
65. Jeon S, Chien J, Song C, Hong J. A preliminary study on precision image the spatiotemporal dynamics of task-initiated thoughts with EEG and fMRI.
guidance for electrode placement in an EEG study. Brain Topogr. (2018) Neuroimage. (2019). doi: 10.1016/j.neuroimage.2019.03.029. [Epub ahead of
31:174–85. doi: 10.1007/s10548-017-0610-y print].
66. Akalin Acar Z, Makeig S. Effects of forward model errors 86. Hernandez F, Johnson RA. The large-sample behavior of transformations to
on EEG source localization. Brain Topogr. (2013) 26:378–96. normality. J Am Stat Assoc. (1980) 75:855–61.
doi: 10.1007/s10548-012-0274-6 87. Biasiucci A, Franceschiello B, Murray MM. Electroencephalography. Curr
67. Baillet S, Riera JJ, Marin G, Mangin JF, Aubert J, Garnero L. Evaluation Biol. (2019) 29:R80–R85. doi: 10.1016/j.cub.2018.11.052
of inverse methods and head models for EEG source localization 88. He B, Yang L, Wilke C, Yuan H. Electrophysiological imaging of brain
using a human skull phantom. Phys Med Biol. (2001) 46:77–96. activity and connectivity-challenges and opportunities. IEEE Trans Biomed
doi: 10.1088/0031-9155/46/1/306 Eng. (2011) 58:1918–31. doi: 10.1109/TBME.2011.2139210
68. Fuchs M, Wagner M, Kastner J. Development of volume conductor and 89. Michel CM, Murray MM. Towards the utilization of EEG as a brain imaging
source models to localize epileptic foci. J Clin Neurophysiol. (2007) 24:101– tool. NeuroImage. (2012) 61:371–85. doi: 10.1016/j.neuroimage.2011.12.039
19. doi: 10.1097/WNP.0b013e318038fb3e 90. Plummer C, Harvey AS, Cook M. EEG source localization in
69. Guggisberg AG, Dalal SS, Zumer JM, Wong DD, Dubovik S, focal epilepsy: where are we now? Epilepsia. (2008) 49:201–18.
Michel CM, et al. Localization of cortico-peripheral coherence doi: 10.1111/j.1528-1167.2007.01381.x
with electroencephalography. NeuroImage. (2011) 57:1348–57. 91. Feng R, Hu J, Pan L, Wu J, Lang L, Jiang S, et al. Application of 256-
doi: 10.1016/j.neuroimage.2011.05.076 channel dense array electroencephalographic source imaging in presurgical
70. Wang G, Worrell G, Yang L, Wilke C, He B. Interictal spike analysis of workup of temporal lobe epilepsy. Clin Neurophysiol. (2015) 127:108–16
high-density EEG in patients with partial epilepsy. Clin Neurophysiol. (2011) doi: 10.1016/j.clinph.2015.03.009
122:1098–105. doi: 10.1016/j.clinph.2010.10.043 92. Klamer S, Elshahabi A, Lerche H, Braun C, Erb M, Scheffler K, et al.
71. Birot G, Spinelli L, Vulliemoz S, Megevand P, Brunet D, Seeck M, et al. Differences between MEG and high-density EEG source localizations using
Head model and electrical source imaging: A study of 38 epileptic patients. a distributed source model in comparison to fMRI. Brain Topogr. (2015)
Neuroimage Clin. (2014) 5:77–83. doi: 10.1016/j.nicl.2014.06.005 28:87–94. doi: 10.1007/s10548-014-0405-3
72. Lillie EM, Urban JE, Lynch SK, Weaver AA, Stitzel JD. Evaluation of skull 93. Lascano AM, Perneger T, Vulliemoz S, Spinelli L, Garibotto V, Korff CM,
cortical thickness changes with age and sex from computed tomography et al. Yield of , high-density electric source imaging (HD-), SPECT and
scans. J Bone Miner Res. (2016) 31:299–307. doi: 10.1002/jbmr.2613 PET in epilepsy surgery candidates. Clin Neurophysiol. (2016) 125:150–5.
73. Roche AF. Increase in cranial thickness during growth. Hum Biol. doi: 10.1016/j.clinph.2015.03.025
(1953) 25:81–92. 94. Chowdhury RA, Merlet I, Birot G, Kobayashi E, Nica A, Biraben A, et al.
74. Goncalves S, de Munck JC, Verbunt JP, Heethaar RM, da Silva FH. In vivo Complex patterns of spatially extended generators of epileptic activity:
measurement of the brain and skull resistivities using an EIT-based method Comparison of source localization methods cMEM and 4-ExSo-MUSIC
and the combined analysis of SEF/SEP data. IEEE Trans Biomed Eng. (2003) on high resolution EEG and MEG data. NeuroImage. (2016) 143:175–95.
50:1124–8. doi: 10.1109/TBME.2003.812164 doi: 10.1016/j.neuroimage.2016.08.044
75. Lai Y, van Drongelen W, Ding L, Hecox KE, Towle VL, Frim DM, 95. Hassan M, Mheich A, Biraben A, Merlet I, Wendling F. Identification
et al. Estimation of in vivo human brain-to-skull conductivity ratio from of epileptogenic networks from dense EEG: A model-based
simultaneous extra- and intra-cranial electrical potential recordings. Clin study. Conf Proc IEEE Eng Med Biol Soc. (2015) 2015:5610–3.
Neurophysiol. (2005) 116:456–65. doi: 10.1016/j.clinph.2004.08.017 doi: 10.1109/EMBC.2015.7319664
76. Oostendorp TF, Delbeke J, Stegeman DF. The conductivity of the human 96. Michel CM, Thut G, Morand S, Khateb A, Pegna AJ, Grave de Peralta R, et al.
skull: results of in vivo and in vitro measurements. IEEE Trans Biomed Eng. Electric source imaging of human brain functions. Brain Res Brain Res Rev.
(2000) 47:1487–92. doi: 10.1109/TBME.2000.880100 (2001) 36:108–118. doi: 10.1016/S0165-0173(01)00086-8
77. Rush S, Driscoll DA. EEG electrode sensitivity–an application of reciprocity. 97. Astolfi L, de Vico Fallani F, Cincotti F, Mattia D, Marciani MG, Bufalari S,
IEEE Trans Biomed Eng. (1969) 16:15–22. doi: 10.1109/TBME.1969.4502645 et al. Imaging functional brain connectivity patterns from high-resolution
78. Hoekema R, Wieneke GH, Leijten FS, van Veelen CW, van Rijen EEG and fMRI via graph theory. Psychophysiology. (2007) 44:880–893.
PC, Huiskamp GJ, et al. Measurement of the conductivity of skull, doi: 10.1111/j.1469-8986.2007.00556.x

Frontiers in Neurology | www.frontiersin.org 17 April 2019 | Volume 10 | Article 325

Michel and Brunet EEG Source Imaging

98. Coito A, Michel CM, van Mierlo P, Vulliemoz S, Plomp G. Directed 106. Chella F, Pizzella V, Zappasodi F, Marzetti L. Impact of the reference choice
functional brain connectivity based on EEG source imaging: methodology on scalp EEG connectivity estimation. J Neural Eng. (2016) 13:036016.
and application to temporal lobe epilepsy. IEEE Trans Biomed Eng. (2016) doi: 10.1088/1741-2560/13/3/036016
63:2619–28. doi: 10.1109/TBME.2016.2619665 107. Engel AK, Gerloff C, Hilgetag CC, Nolte G. Intrinsic coupling modes:
99. He B, Dai Y, Astolfi L, Babiloni F, Yuan H, Yang L. eConnectome: A multiscale interactions in ongoing brain activity. Neuron. (2013) 80:867–86.
MATLAB toolbox for mapping and imaging of brain functional connectivity. doi: 10.1016/j.neuron.2013.09.038
J. Neurosci. Methods. (2011) 195:261–9. doi: 10.1016/j.jneumeth.2010. 108. Haufe S, Nikulin VV, Muller KR, Nolte G. A critical assessment of
11.015 connectivity measures for EEG data: a simulation study. NeuroImage. (2013)
100. Astolfi L, De Vico Fallani F, Cincotti F, Mattia D, Marciani MG, Salinari 64:120–33. doi: 10.1016/j.neuroimage.2012.09.036
S, et al. Estimation of effective and functional cortical connectivity from 109. Schoffelen JM, Gross J. Source connectivity analysis with MEG and EEG.
neuroelectric and hemodynamic recordings. IEEE Trans Neural Syst Rehabil Hum Brain Mapp. (2009) 30:1857–65. doi: 10.1002/hbm.20745
Eng. (2009) 17:224–33. doi: 10.1109/TNSRE.2008.2010472 110. Van de Steen F, Faes L, Karahan E, Songsiri J, Valdes-Sosa PA, Marinazzo
101. Coito A, Michel CM, Vulliemoz S, Plomp G. Directed functional connections D. Critical comments on EEG sensor space dynamical connectivity analysis.
underlying spontaneous brain activity. Hum Brain Mapp. (2019) 40:879–88. Brain Topography. (2016). doi: 10.1007/s10548-016-0538-7
doi: 10.1002/hbm.24418 111. He B, Astolfi L, Valdés-Sosa PA, Marinazzo D, Palva SO, Bénar CG, et al.
102. Wendling F, Ansari-Asl K, Bartolomei F, Senhadji L. From EEG signals to Electrophysiological brain connectivity: theory and applications. IEEE Trans
brain connectivity: a model-based evaluation of interdependence measures. Biomed Eng. (2019).
J Neurosci Methods. (2009) 183:9–18. doi: 10.1016/j.jneumeth.2009.04.021
103. Coito A, Genetti M, Pittau F, Iannotti G, Thomschewski A, Höller Y, et al. Conflict of Interest Statement: The authors declare that the research was
Altered directed connectivity in temporal lobe epilepsy in the absence of conducted in the absence of any commercial or financial relationships that could
interictal spikes: a high density EEG study. Epilepsia. (2016) 57:402–11. be construed as a potential conflict of interest.
doi: 10.1111/epi.13308
104. Coito A, Plomp G, Genetti M, Abela E, Wiest R, Seeck M, et al. Dynamic Copyright © 2019 Michel and Brunet. This is an open-access article distributed
directed interictal connectivity in left and right temporal lobe epilepsy. under the terms of the Creative Commons Attribution License (CC BY). The use,
Epilepsia. (2015) 56:207–17. doi: 10.1111/epi.12904 distribution or reproduction in other forums is permitted, provided the original
105. Brunner C, Billinger M, Seeber M, Mullen TR, Makeig S. Volume conduction author(s) and the copyright owner(s) are credited and that the original publication
influences scalp-based connectivity estimates. Front Comput Neurosci. in this journal is cited, in accordance with accepted academic practice. No use,
(2016) 10:121. doi: 10.3389/fncom.2016.00121 distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Neurology | www.frontiersin.org 18 April 2019 | Volume 10 | Article 325