ArticlePDF Available

Abstract

Children differ from adults in head size, skull morphology, and tissue conductivity. We conducted a simulation to examine the error of source localization when a rescaled adult head model and different skull conductivities are used for EEG source localization in children. We have proven by simulation that source localization accuracy is the best with an infant specific head model including the age specific skull structure and conductivity.
Anatomically Accurate Infant Head Models for EEG Source Localization
This content has been downloaded from IOPscience. Please scroll down to see the full text.
Download details:
IP Address: 128.252.175.159
This content was downloaded on 19/09/2013 at 21:15
Please note that terms and conditions apply.
2013 J. Phys.: Conf. Ser. 434 012012
(http://iopscience.iop.org/1742-6596/434/1/012012)
View the table of contents for this issue, or go to the journal homepage for more
Home Search Collections Journals About Contact us My IOPscience
Anatomically Accurate Infant Head Models for EEG
Source Localization
Jasmine Song1, Sergei Turovets1,2, Pavel Govyadinov1,2, Chelsea
Mattson1, Phan Luu1,3, Kirk Smith4, Fred Prior4, Linda
Larson-Prior4and Don M. Tucker1,3
1Electrical Geodesics, Inc. 1600 Millrace Dr. Suite 200, Eugene OR 97403, USA
2Neuroinformatics Center, University of Oregon, Eugene OR 97403, USA
3Department of Psychology, University of Oregon, Eugene OR 97403, USA
4Electrical and Optical Imaging, Neuroimaging Laboratory, Washington University, St. Louis,
MO, USA
E-mail: jsong@egi.com
Abstract. Children differ from adults in head size, skull morphology, and tissue conductivity.
We conducted a simulation to examine the error of source localization when a rescaled adult
head model and different skull conductivities are used for EEG source localization in children.
We have proven by simulation that source localization accuracy is the best with an infant specific
head model including the age specific skull structure and conductivity.
1. Introduction
Dense array electroencephalography (dEEG) is an important noninvasive imaging modality for
elucidating normal cognitive development in children and brain monitoring in neonates intensive
care units. At present, the use and validation of dEEG and Electromagnetic Source Localization
(ESL) in infants and children is hindered by the lack of pediatric head models. To generate an
ESL solution, two independently specified problems must be solved: (1) the forward problem
which specifies how currents move from their site of generation in cortex to the scalp, and (2) the
inverse problem which is highly underspecified, and works from the recorded scalp potentials to
the source of those signals. While many studies have shown that properly constrained realistic
head models based on the high resolution MRI generate more accurate ESL solutions, the
computability and necessity of additional variables such as tissue inhomogeneity and anisotropy
are less well understood. While this is true of the current head models for normal adult
populations, it is even more critical for pediatric populations where both the shape and density of
the skull and the volume and composition of the brain are changing rapidly during first few years
of life. Therefore, accurate ESL requires an accurate lead field matrix (LFM), specifying the
forward lead fields, from each cortical source to each head surface electrode. Children differ from
adults in head geometry, skull thickness, and tissue conductivity. We conducted a simulation
to examine the error when an adult skull model and different skull conductivities are used in
pediatric ESL. Specifically, we analyze three pediatric head models based on 1) a 7 months old
infant MRI coregistered with the same subject CT, 2) the same MRI with warped adult CT
atlas skull and adjusted thickness, and 3) an adult head rescaled to the 7 month old infant size.
XV Int. Conf. on Electrical Bio-Impedance & XIV Conf. on Electrical Impedance Tomography IOP Publishing
Journal of Physics: Conference Series 434 (2013) 012012 doi:10.1088/1742-6596/434/1/012012
Published under licence by IOP Publishing Ltd
1
The effects of the geometry (like presence or absence of fontanelles) and the skull conductivity
specifications were examined. All three models were analyzed for six conductivity values ranging
from the lowest value reported in the literature [1] 0.004 S/m through the average adult value
0.018 S/m to the average scalp conductivity (effectively no skull) and 18 (3 ×6) LFMs have
been generated by our in-house finite difference forward solver for ESL. We have chosen a set
of representative dipoles near fontanelles on the infant cortical surface in the model as the
“synthetic ground truth” EEG data assuming 0.1 S/m to be the true infant skull conductivity
value, and localized the sources with the minimum norm (MN) and sLORETA [2] distributed
source localization methods using the rest of LFMs for 3 geometries with 6 skull conductivity
specifications.
2. Methods
2.1. Forward problem
MRI image segmentation and registration were conducted with the EGI MRI/CT software
package, BrainK[3] that is capable of cortical surface extraction and skull data warping to subject
specific head shapes obtained from MRI scans. Sensor positions and head shapes are provided
by the Geodesic Photogrammetry System (GPS) [4]. For the ground truth and reference, a true
infant model(True) was created using subject specific MRI/CT data acquired at Washington
University [5] and sensor positions for a 7 month-old baby (figure 1). After the segmentation and
cortex tessellation, the oriented dipoles grid was constructed and LFMs were generated with our
in-house finite difference solver [6]. We calculated 18 different LFMs, comparing 3 geometries and
6 skull conductivities. For comparison an adult skull was warped (Warped) to match subject
specific MRI and an adult model (Adult) was simply rescaled to match the pediatric sensor
positions, retaining the adult tissue morphology. In both cases the fontanelle structure was lost
(figure 2). The ground truth EEG was calculated assuming a uniform skull conductivity of 0.1
S/m and the true infant model for several dipole locations: near the fontanelles, the eyeball and
deep in the brain. The skull conductivities were varied according to the set of values 0.004 S/m,
0.018 S/m, 0.1 S/m, 0.2 S/m, 0.35 S/m and 0.45 S/m.
(a) Sensors on Scalp (b) Segmentation (c) Cortical surface
Figure 1. Sensors on pediatric head were registered by geodesic photogrammetry. A relative
thresholding method segmented the scalp, skull, CSF, and gray and white matter. The cortical
surface was extracted.
(a) True (b) Warped (c) Adult
Figure 2. Skulls of three geometries. The true model contains frontal and occipital fontanelles.
Both warped and adult head models don’t have fontanelles.
XV Int. Conf. on Electrical Bio-Impedance & XIV Conf. on Electrical Impedance Tomography IOP Publishing
Journal of Physics: Conference Series 434 (2013) 012012 doi:10.1088/1742-6596/434/1/012012
2
2.2. Inverse problem
We use the MN and sLORETA methods of ESL. The distributed source localization can be
stated as following: for measured Φ and known K, find Jgiven Φ = KJ , where Φ is the electric
potential at the electrodes of size Ne,Jis the (unknown) amplitude of each current dipole of
size Nv, and Kis the lead field matrix linking the current sources to the electric potential of size
Ne×Nv. The MN solution is ˆ
J= arg minJ{kΦKJk2+αkJk2}, where αis a non-negative
regularization parameter and k · k2represents the square of the l2-norm. The MN solution ˆ
Jis
ˆ
J=KT[KKT+αINe]1Φ. The sLORETA solution ˆ
Jis ˆ
J
l= [Cˆ
J]1/2
ll ˆ
Jl, where l= 1, . . . , Nv
and Cˆ
J=KT[KKT+αINe]1K.
After the inverse problems are solved, the source solutions are validated by estimating the
localization error distance (LED). The LED is the Euclidean distance between the locations
of true dipole and the dipole with maximum intensity in the source distribution. To separate
the effect of conductivity variation from differences in the geometry on source estimates, the
ground truth EEG was calculated in each model assuming a uniform skull conductivity of 0.1
S/m. The sources were localized using all 6 LFMs with different conductivities. To see the
effect of the head geometry, we have focused on a dipole near the rear fontanelles in the subject
specific model with the skull conductivity of 0.1 S/m. The forward projection of the dipole was
considered as the true EEG. We performed the source localization of this EEG data with the 3
LFMs created for 3 different geometries with skull conductivity 0.1 S/m.
3. Results
Table 1 summarizes the LEDs to assess the effect of skull conductivity. The LED is minimal,
for both MN and sLORETA solutions, when the conductivity values used in the head model
matches with those used to generate the synthetic EEG data. Use of the head models with the
correct skull conductivity values results in the minimal LEDs. When skull conductivity values
are small (skull is more resistive), the errors are larger.
The effect of head geometry on source localization can be seen in figure 3. A dipole was
placed near the frontal fontanelle in the true model with skull conductivity of 0.1 S/m. Scalp
potentials of the forward projections of the chosen dipole shows that the potential distribution
is very focal. Since the dipole is located near the head surface, it can be easily localized. LEDs
for true and warped head models are 0 mm but 5.39 mm for adult head model in both the MN
and sLORETA methods. Placing the dipoles in other brain regions we showed that accurate
head geometries are significantly more important for the deep sources than the shallow ones.
Table 1. The mean of LED is in mm. LFMs with 0.1 S/m obtain the minimum LED in three
geometries by both MN and sLORETA.
Method Head Geometry Skull Conductivity (S/m)
0.004 0.018 0.10.2 0.35 0.45
MN True 17.3 15.5 15.1 15.1 15.2 15.2
Warped 18.4 16.2 15.2 15.3 15.4 15.5
Adult 19.8 18.3 13.9 14.9 16.3 16.6
sLORETA True 3.7 0.3 0 0 0 0.0
Warped 7.6 1.1 0 0 0.0 0.1
Adult 10.6 4.6 0 0.1 0.5 1.0
XV Int. Conf. on Electrical Bio-Impedance & XIV Conf. on Electrical Impedance Tomography IOP Publishing
Journal of Physics: Conference Series 434 (2013) 012012 doi:10.1088/1742-6596/434/1/012012
3
Dipole Scalp MN sLORETA
(a) True
(b) Warped
(c) Adult
Figure 3. From left to right: a probe dipole near the frontal fontanelle in the true model is
placed at the same location in both comparison cases; scalp potential of the forward projections
of the probe dipole; localized sources using MN and sLORETA.
4. Conclusions
Accurate ESL with infants and young children requires electrical lead fields (head models)
constructed to match both the unique geometries and skull conductivities of the appropriate
age child instead of using rescaled adult head models. The actual conductivity values for infant
skull are not well known presently, but unlikely to be in the adult range of low values. Therefore
further studies are needed and methods like bounded EIT [6] can be used for noninvasive regional
tissue conductivity estimates in infants.
Acknowledgments
This study was supported by National Institute of Health grant number R43NS067726-02.
References
[1] Roche-Labarbe N, Aarabi A, Kongolo G, Gondry-Jouet C, Dumpelmann M, Grebe R and
Wallois F 2008 Human Brain Mapping 29 167–176
[2] Pascual-Marqui R D 2002 Methods and Findings in Experimental and Clinical
Pharmacology 24 5–12
[3] Li K, Malony A and Tucker D M 2006 First International Conference on Computer Vision
Theory and Applications (VISAPP), Portugal, February pp 354–364
[4] Russell G S, Eriksen K J, Poolman P, Luu P and Tucker D M 2005 Clinical Neurophysiology
116 1130–40
[5] Larson-Prior L, Prior F, Smith K, Mattson C, Song J, Nolan T S, Luu P, Tucker D and
Turovets S 2013 IEEE Explore, EMBC, June 2013
[6] Salman A, Turovets S, Malony A and Volkov V 2008 International Workshop on OpenMP
(IWOMP 2005/2006) LNCS 4315 pp 119–130
XV Int. Conf. on Electrical Bio-Impedance & XIV Conf. on Electrical Impedance Tomography IOP Publishing
Journal of Physics: Conference Series 434 (2013) 012012 doi:10.1088/1742-6596/434/1/012012
4
... Lastly, the standard MNI brain used in the sLORETA software is not derived from the pediatric population, which differs from adults in head size, skull shape, and tissue conductivity [39]. However, head models of infants or neonates are not frequently used currently. ...
Article
Full-text available
Background It is important to assess the degree of brain injury and predict long-term outcomes in neonates diagnosed with hypoxic-ischemic encephalopathy (HIE). However, routine studies, including magnetic resonance imaging (MRI) and conventional encephalography (EEG) or amplitude-integrated EEG (aEEG), have their own limitations in terms of availability and accuracy of evaluation. Recently, quantitative EEG (qEEG) has been shown to improve the predictive reliability of neonatal HIE and has been further refined with brain mapping techniques. Methods We investigated background EEG activities in 29 neonates with HIE who experienced therapeutic hypothermia, via qEEG using a distributed source model. MRI images were evaluated and classified into two groups (normal-to-mild injury vs moderate-to-severe injury), based on a scoring system. Non-parametric statistical analysis using standardized low-resolution brain electromagnetic tomography was performed to compare the current density distribution of four frequency bands (delta, theta, alpha, and beta) between the two groups. Results Electrical neuronal activities were significantly lower in the moderate-to-severe injury group compared with the normal-to-mild injury group. Background EEG activities in moderate-to-severe HIE were most significantly reduced in the temporal and parietal lobes. Quantitative EEG also revealed a decrease in background activity at all frequency bands, with a maximum in decrease in the delta component. The maximum difference in current density was found in the inferior parietal lobule of the right parietal lobe for the delta frequency band. Conclusions Our study demonstrated quantitative and topographical changes in EEG in moderate-to-severe neonatal HIE. They also suggest possible implementation and evaluation of conventional EEG and aEEG in neonatal HIE. The findings have implications as biomarkers in the assessment of neonatal HIE.
... Note: The forward problem involves the definition of a head model that represents the geometry of the head and of the electrical conductivity properties of the head volume 9,10 . ...
Article
Full-text available
Understanding typical and atypical development remains one of the fundamental questions in developmental human neuroscience. Traditionally, experimental paradigms and analysis tools have been limited to constrained laboratory tasks and contexts due to technical limitations imposed by the available set of measuring and analysis techniques and the age of the subjects. These limitations severely limit the study of developmental neural dynamics and associated neural networks engaged in cognition, perception and action in infants performing "in action and in context". This protocol presents a novel approach to study infants and young children as they freely organize their own behavior, and its consequences in a complex, partly unpredictable and highly dynamic environment. The proposed methodology integrates synchronized high-density active scalp electroencephalography (EEG), inertial measurement units (IMUs), video recording and behavioral analysis to capture brain activity and movement non-invasively in freely-behaving infants. This setup allows for the study of neural network dynamics in the developing brain, in action and context, as these networks are recruited during goal-oriented, exploration and social interaction tasks.
... The results have to be interpreted very cautiously, particularly given that substantial differences in head size, skull morphology, and tissue conductivity are found between adults and infants. Nevertheless, they are expected to convey at least approximate anatomical information, particularly given that while reduced head models provide better source localization accuracy for infants, the error associated with usage of adult head models does not appear to be such as to invalidate lobar-level inferences (Song et al., 2013). Furthermore, seemingly consistent localization patterns are found in studies where adults and children are compared while using a standard head model (Thatcher, North, & Biver, 2007). ...
Article
Full-text available
People react aversely when faced with unfair situations, a phenomenon that has been related to an electroencephalographic (EEG) potential known as medial frontal negativity (MFN). To our knowledge, the existence of the MFN in children has not yet been demonstrated. Here, we recorded EEG activity from 15 children playing the ultimatum game (UG) and who afterward performed a recognition task, in order to assess whether they could recognize the unfair and fair (familiar) proposers among unfamiliar faces. During the recognition task, we also acquired pupil dilation data to investigate subconscious recognition processes. A typical (adult-like) MFN component was detected in reaction to unfair proposals. We found a positive correlation between reaction time and empathy, as well as a negative correlation between reaction time and systematic reasoning scores. Finally, we detected a significant difference in pupil dilation in response to unfamiliar faces versus UG proposers. Our data provide the first evidence of MFN in children, which appears to index similar neurophysiological phenomena as in adults. Also, reaction time to fair proposals seems to be related to individual traits, as represented by empathy and systematizing. Our pupil dilation data provide evidence that automatic responses to faces did not index fairness, but familiarity. These findings have implications for our understanding of social development in typically developing children.
Article
Full-text available
Electromagnetic source characterisation requires accurate volume conductor models representing head geometry and the electrical conductivity field. Head tissue conductivity is often assumed from previous literature, however, despite extensive research, measurements are inconsistent. A meta-analysis of reported human head electrical conductivity values was therefore conducted to determine significant variation and subsequent influential factors. Of 3121 identified publications spanning three databases, 56 papers were included in data extraction. Conductivity values were categorised according to tissue type, and recorded alongside methodology, measurement condition, current frequency, tissue temperature, participant pathology and age. We found variation in electrical conductivity of the whole-skull, the spongiform layer of the skull, isotropic, perpendicularly- and parallelly-oriented white matter (WM) and the brain-to-skull-conductivity ratio (BSCR) could be significantly attributed to a combination of differences in methodology and demographics. This large variation should be acknowledged, and care should be taken when creating volume conductor models, ideally constructing them on an individual basis, rather than assuming them from the literature. When personalised models are unavailable, it is suggested weighted average means from the current meta-analysis are used. Assigning conductivity as: 0.41 S/m for the scalp, 0.02 S/m for the whole skull, or when better modelled as a three-layer skull 0.048 S/m for the spongiform layer, 0.007 S/m for the inner compact and 0.005 S/m for the outer compact, as well as 1.71 S/m for the CSF, 0.47 S/m for the grey matter, 0.22 S/m for WM and 50.4 for the BSCR.
Preprint
Electromagnetic source characterisation requires accurate volume conductor models representing head geometry and the electrical conductivity field. Head tissue conductivity is often assumed from previous literature, however, despite extensive research, measurements are inconsistent. A meta-analysis of reported human head electrical conductivity values was therefore conducted to determine significant variation and subsequent influential factors. Of 3,121 identified publications spanning three databases, 56 papers were included in data extraction. Conductivity values were categorised according to tissue type, and recorded alongside methodology, measurement condition, current frequency, tissue temperature, participant pathology and age. We found variation in electrical conductivity of the whole-skull, the spongiform layer of the skull, isotropic, perpendicularly- and parallelly-oriented white matter (WM) and the brain-to-skull-conductivity ratio (BSCR) could be significantly attributed to a combination of differences in methodology and demographics. Specifically, whole-skull conductivity varied significantly with the employed method, whilst the spongiform layer differed with measurement condition. Grey matter conductivity fluctuated according to method and pathology, whilst isotropic WM differed according to all of the independent variables. Finally, the BSCR significantly varied with participants age. The reported results provided evidence for head conductivity inconsistencies being dependent on methodology and demographics, supporting the hypothesis that conductivity values should not be assumed from the literature.
Article
Full-text available
Background: Electroencephalography (EEG) is a popular method to monitor brain activity, but it is difficult to evaluate EEG-based analysis methods because no ground-truth brain activity is available for comparison. Therefore, in order to test and evaluate such methods, researchers often use simulated EEG data instead of actual EEG recordings. Simulated data can be used, among other things, to assess or compare signal processing and machine learning algorithms, to model EEG variabilities, and to design source reconstruction methods. New method: We present SEREEGA, Simulating Event-Related EEG Activity. SEREEGA is a free and open-source MATLAB-based toolbox dedicated to the generation of simulated epochs of EEG data. It is modular and extensible, at initial release supporting five different publicly available head models and capable of simulating multiple different types of signals mimicking brain activity. This paper presents the architecture and general workflow of this toolbox, as well as a simulated data set demonstrating some of its functions. The toolbox is available at https://github.com/lrkrol/SEREEGA. Results: The simulated data allows established analysis pipelines and classification methods to be applied and is capable of producing realistic results. Comparison with existing methods: Most simulated EEG is coded from scratch. The few open-source methods in existence focus on specific applications or signal types, such as connectivity. SEREEGA unifies the majority of past simulation methods reported in the literature into one toolbox. Conclusion: SEREEGA is a general-purpose toolbox to simulate ground-truth EEG data.
Preprint
Full-text available
Electroencephalography (EEG) is a popular method to monitor brain activity, but it can be difficult to evaluate EEG-based analysis methods because no ground-truth brain activity is available for comparison. Therefore, in order to test and evaluate such methods, researchers often use simulated EEG data instead of actual EEG recordings, ensuring that it is known beforehand which e ects are present in the data. As such, simulated data can be used, among other things, to assess or compare signal processing and machine learn-ing algorithms, to model EEG variabilities, and to design source reconstruction methods. In this paper, we present SEREEGA, short for Simulating Event-Related EEG Activity . SEREEGA is a MATLAB-based open-source toolbox dedicated to the generation of sim-ulated epochs of EEG data. It is modular and extensible, at initial release supporting ve different publicly available head models and capable of simulating multiple different types of signals mimicking brain activity. This paper presents the architecture and general work ow of this toolbox, as well as a simulated data set demonstrating some of its functions. Highlights Simulated EEG data has a known ground truth, which can be used to validate methods. We present a general-purpose open-source toolbox to simulate EEG data. It provides a single framework to simulate many different types of EEG recordings. It is modular, extensible, and already includes a number of head models and signals. It supports noise, oscillations, event-related potentials, connectivity, and more.
Article
Full-text available
Although Electroencephalography (EEG) source localization is being widely used in adults, this promising technique has not yet been applied to newborns because of technical difficulties, such as lack of data concerning the newborn skull conductivity, thickness, and homogeneity. Using a new type of EEG headcap molded on each baby's head, we aimed to determine whether this technique could be adapted to neonates, and to evaluate the importance of these technical difficulties. We carried out EEG source reconstruction of the recordings of five neonates using dipole fit algorithm. We used four different head models for each neonate, obtained from individual MRI scans: normal skull thickness and conductivity of 0.0042 S/m; normal thickness and conductivity of 0.33 S/m; increased thickness and conductivity of 0.0042 S/m; and normal thickness and conductivity with a modeled bregma fontanel. Dipole locations were consistent with MRI and clinical data. The mean difference between the dipole locations in the 0.0042 and the 0.33 S/m skull layer models was 11.6 +/- 2.5 mm, with an average 29.7% decrease in magnitude for the 0.33 S/m model but no significant changes for the dipoles orientation. Skull layer thickness had a large influence on magnitude, but no significant effect on position and orientation. The mean difference between the dipole locations induced by the modeled fontanel was 2.0 +/- 2.1 mm, with an average 2.1% increase in magnitude. Our results show that EEG source localization is feasible in neonates. With further development, the technique may prove useful for neurological evaluation of neonates.
Article
An important goal for functional brain studies using EEG technology is to estimate the location of brain sources that produce the scalp-recorded signals. The accuracy of source estimates is dependent upon many variables, one of which is the accurate description of the scalp positions of the EEG sensors. The objective of the present research was to develop a photogrammatic method for sensor localization that is fast, accurate, and easy to use. With the novel photogrammetric method, multiple cameras were arranged in a geodesic array, and images of the sensors on the subject's head were acquired allowing for the reconstruction of the 3D sensor positions. Data from the photogrammetric method were compared with data acquired with the conventional electromagnetic method. The accuracy of the photogrammatic method, quantified as RMS of the measured positions and the actual known positions, was similar (mean error = 1.27 mm) to the electromagnetic method (mean error = 1.02 mm), and both approximated the localization error of the calibration object (mean error = 0.56 mm). Accurate determination of 3D sensor positions can be accomplished with minimal demands on the time of the subject and the experimenter using the photogrammetric method.
  • G S Russell
  • K J Eriksen
  • P Poolman
  • P Luu
  • D Tucker
Russell G S, Eriksen K J, Poolman P, Luu P and Tucker D M 2005 Clinical Neurophysiology 116 1130–40
  • A Salman
  • S Turovets
  • A Malony
  • V Volkov
Salman A, Turovets S, Malony A and Volkov V 2008 International Workshop on OpenMP (IWOMP 2005/2006) LNCS 4315 pp 119–130
  • R Pascual-Marqui
Pascual-Marqui R D 2002 Methods and Findings in Experimental and Clinical Pharmacology 24 5–12
  • N Roche-Labarbe
  • A Aarabi
  • G Kongolo
  • C Gondry-Jouet
  • M Dumpelmann
  • R Grebe
  • F Wallois
Roche-Labarbe N, Aarabi A, Kongolo G, Gondry-Jouet C, Dumpelmann M, Grebe R and Wallois F 2008 Human Brain Mapping 29 167–176