Upload
mrh12345
View
222
Download
0
Embed Size (px)
Citation preview
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 1/7
Variability in spatial normalization of pediatric and adult brain images
Marco R. Hoeksmaa,*, J. Leon Kenemansb, Chantal Kemnerc, Herman van Engelandc
a Department of Psychopharmacology, Faculty of Pharmaceutical Sciences, Utrecht University, Sorbonnelaan 16, 3584CA Utrecht, The Netherlandsb Department of Psychonomics, Faculty of Social Sciences, Utrecht University, Utrecht, The Netherlands
c Department of Child- and Adolescent Psychiatry, Utrecht University Medical Center, Utrecht, The Netherlands
Accepted 21 December 2004
Abstract
Objective: Normalization of brain images is a necessity for group comparisons of source analyses based on realistic head models. In thispaper we compared the outcome of a linear registration method for brain images of psychiatric and control groups of different ages in order to
assess the relative adequacy of normalization in such diverse groups.
Methods: Magnetic Resonance images (MRI) of the brains of pediatric and adolescent subjects (mean ages 19 and 10.5 years) with a
pervasive developmental disorder (PDD) and their healthy controls were included. A simple voxel-wise test of the group variances in image
intensities was performed to evaluate regional differences in registration quality. Dipole analysis of visual P1 was performed to establish
whether source locations were comparable across groups.
Results: Significant differences between pediatric groups were found in white matter and thalamic regions of the brain. For all other group-
wise comparisons, differences were confined to skull and neck regions. Dipole locations were found to be more anteriorly located in the
adolescent groups.
Conclusions: The normalization procedure used in this paper is based on a brain template of normal adult brains from a restricted age
group, and the results show that the use of this method in pediatric groups is less adequate. The method seems suitable for use in psychiatric
groups. Also, the generators of visual P1 in PDD patients were found to be comparable to controls.
Significance: The results suggest that this existing normalization method can be used in diverse populations, but is less suitable forpediatric images.
q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: Head model; Magnetic resonance imaging; Source localization; Spatial normalization; Pervasive developmental disorder; P1
1. Introduction
In order to increase the accuracy of head models for
electrical and magnetic source analyses, structural brain
images have entered the field of electrophysiology (Fuchs
et al., 2001; Wagner, 1998). Traditionally, spherical models
were used to model the head (Scherg, 1990). The assumedspherical dimensions in these models were similar for all
subjects and a direct link to the anatomy was missing.
Structural images of the head have the advantage of
enabling the researcher to construct a head model that is
tailor made for each subject under study, thus improving
localization accuracy and opening the way for inferences
about the brain structures involved in the modeled
electrophysiological phenomena.
Although individual head models improve the quality
of single-subject localizations, they introduce a new
source of variance in group-wise comparisons of localiz-
ation results. In order to facilitate group comparisons and
the generalization of findings, individual variability inbrain images has to be minimized. This minimization, or
spatial normalization, is achieved by transforming the
individual image to a common template that is aligned in
standardized stereotaxic space. However, it may be
that the use of such a common standardized template
results in errors that differ in size or nature across age- or
patient groups. The present work addresses potential
differences in error patterns between typical age and
clinical groups.
Clinical Neurophysiology 116 (2005) 1188–1194
www.elsevier.com/locate/clinph
1388-2457/$30.00 q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.clinph.2004.12.021
* Corresponding author. Tel.:C31 30 2533845; fax: C31 30 2537387.
E-mail address: [email protected] (M.R. Hoeksma).
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 2/7
Spatial normalization procedures broadly fall into two
categories: intrinsic and extrinsic registration approaches
(Maintz and Viergever, 1998). Extrinsic techniques rely on
external landmarks or fiducial markers that are identified in
both the template and the image to be transformed. The
reliability of such techniques is completely dependent upon
the reproducibility of the markers, which often have to bedefined by means of an interactive procedure. Intrinsic
techniques on the other hand use information that is already
present in the image, either from segmented objects or from
the entire image. Image registration then entails the
optimization of some index of the difference between the
source and the template image, such as the minimization of
the distance between homologous external markers or the
maximization of cross-correlations between voxel values.
A widely used template for such normalization pro-
cedures is the Talairach template (Talairach and Tournoux,
1988). This system relies on aligning the anterior and
posterior commissures (AC PC) and the interhemispheric
plane of the source image with a template brain. Although
widely accepted in the field of neuroimaging, the original
template brain has a number of shortcomings, compromis-
ing its generalizability to a wider population. Probably the
most important limitation is that the Talairach system is
based on post-mortem sections of a single brain: that of a
60-year-old female. Furthermore, complete brain symmetry
has to be assumed since only one hemisphere is given in
detail. This clearly does not represent the majority of
subjects in in vivo imaging studies (Toga and Thompson,
2001).
Another widely used template from the Montreal
Neurological Institute (MNI) was constructed from 305MRI scans of healthy subjects (239 male, 66 female; mean
age 23.4G4.1 years) that were individually registered to
Talairach space and subsequently averaged (Evans et al.,
1992). The use of this template in automated registration
methods is now widespread. Although this template is
intuitively better suited to serve as a template for normal-
ization, it still remains to be established whether it is
applicable to data from subjects that do not fit in the specific
age-range or from subjects with neuropsychiatric conditions
that may affect brain anatomy.
In the present study, it was assessed whether the
adequacy of an automatic spatial normalization
procedure (Collins et al., 1994; http://www.bic.mni.mcgill.
ca/software) when used in children and psychiatric patients
differs from that in healthy subjects. The linear registration
procedure studied in this paper aligns the source brains with
the MNI template, and it may well be that this template is
less suited to the normalization of pediatric scans.
Furthermore, there may be structural abnormalities in the
brains of (psychiatric-) patients that may impede successful
registration. Therefore we included patients diagnosed with
a pervasive developmental disorder (PDD), a psychiatric
condition known to have effects on regional brain anatomy
(Courchesne et al., 1993; Piven et al., 1990, 1996) that are
possibly age dependent (Carper et al., 2002; Courchesne
et al., 2001).
It is important to know whether there are any diagnosis-
or age-related differences in the outcome of the registration
procedure; it may be even more important to know where
such deviations occur. The normalization method used in
this paper (Collins et al., 1994) uses cross correlationsbetween source and template brain as an objective function
that is to be maximized by the normalization process; higher
final objective values indicate increased similarity between
source and template. However, a difference in objective
values can be the result of a structural bias in normalization
or of random normalization differences. Two questions can
be raised regarding the success of normalization: (1) are the
individual brains well matched to the template (as is
expressed by the cross correlations) and (2) are the
individual brains well matched to each other. Here, we
propose a measure of voxel variance to answer the second
question for comparisons between different groups of
subjects. Successful normalization should maximize the
similarity between images, and thus corresponding voxels
should have similar values across individual images.
Consequently, the voxelwise variance within a group
reflects the success of normalization. The extent to which
groups differ in the voxelwise or average (across voxels)
variance can be tested statistically. A relatively low
objective value, paired with low voxel variances within a
group thus points in the direction of a structural bias in
normalization, whereas a low objective value paired with
high variances would indicate random differences. The
spatial distribution of significant differences in voxel-
variance can be represented graphically, and indicateswhere significant differences arise between groups. Such a
representation may identify areas with significant differ-
ences in voxel variance when the overall variance (across
voxels) does not differ.
To demonstrate the application of the spatial normal-
ization procedure in a source localization procedure in
diverse groups, source localizations were performed on
visual P1. This, however, cannot be seen as a litmus test for
the success of the normalization procedure, since age-
related changes in P1 generators cannot be ruled out
(Allison et al., 1984; Levine et al., 2000; Onofrj et al.,
2001). Moreover, although no abnormalities have been
reported in visual P1 in patients with PDD, this does not
imply that none exist. In the present study, we expected
more anteriorly located dipoles in the older age group due to
the increased contributions of frontocentral generators
(Dustman et al., 1982), and no effects of diagnosis.
2. Materials and methods
MRI scans were available for four groups of subjects:
adolescent patients with a pervasive developmental disorder
(PDD, nZ13, mean age 19.9G2.9 years) and adolescent
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–1194 1189
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 3/7
healthy subjects (nZ13, mean age 18.2G0.74 years), PDD
children (nZ11, mean age 10.6G1.56 years) and healthy
control children (nZ14, mean age 10.4G1.13 years).
Diagnoses of PDD were made by child psychiatrists and
were corroborated by scores on the Autism Diagnostic
Inventory, revised edition (Lord et al., 1994). All subjects
were free of medication and had no significant neurologicalhistory. IQ was matched within age groups. PDD children
came form the department of Child- and Adolescent
Psychiatry of the Utrecht Academic Hospital. Adolescent
PDD subjects were patients from the Dr Leo Kanner house,
a residential institution for persons with autism. Controls
were recruited from schools in and around Utrecht. All
participants or their parents gave informed consent and the
study adhered to the guidelines of the hospital’s medical
ethical committee. Before entering the MRI scanning
procedure, children were extensively informed and famil-
iarized with the scanning apparatus by means of photo-
graphs, sound recordings of scanner noise and a mock-up of
the head coil. While in the scanner, participants listened to
music of their own choice and were able to see the MRI
operator through a mirror. To limit head movements during
scanning, the head was fixated with foam cushions and a
head band.
Coronal T1 weighted spoiled gradient echo (3D-FFE)
MR images of the whole head were acquired on a 1.5
Tesla Philips Gyroscan NT scanner (1!1!1.5 mm slices
for children, 1!1!1.2 mm voxels for adolescents, 256!
256 matrix, TEZ4.6 ms, TRZ30 ms, flip angle 308,
contiguous slices). Total acquisition time was about
12 min. Volumes were converted to MINC format and
resliced into 1!1!1 mm voxels. Subsequently, anintensity non-uniformity correction was applied to reduce
artifacts due to field inhomogeneities (Sled et al., 1998).
Also, an overall intensity normalization was performed
over all scans. All volumes were then non-interactively
registered with the MNI template brain, using an
automatic linear procedure provided by the MNI (Collins
et al., 1994). The procedure optimizes the image intensity
cross-correlations between template and target images.
Since spatial detail in the MNI template is degraded due
to averaging, the registration procedure includes a number
of intermediate blurring steps for the source image. These
blurring steps use a Gaussian kernel and filter the higher
spatial details out of the source image, thus increasing
similarity with the target template. The individual images
were then transformed with the resulting parameters for
rotation, translation and scaling. After this spatial normal-
ization, the images were averaged per age- and diagnostic
group and the variance for each average was computed,
thus resulting in four average volumes with four
corresponding variance-images. The variance images
express the deviations of voxel intensity values from the
group mean, distributed throughout the entire volume.
The Hartley F -max test (Levy, 1975) was then applied to
the respective groupwise comparisons of voxel values
(adolescent controls vs. young controls, adolescent PDD
vs. young PDD, young PDD vs. young controls and
adolescent PDD vs. adolescent controls). The F -max
statistic is the ratio of the largest variance divided by the
smallest variance. Voxels that differed significantly with
P!0.01 (F -maxZ23.2) were colour-coded for each
comparison. The colour-coded images were then plottedas superimposed coronal sections over the background of
the grand average brain volume for display (spatial extent
K106 through 102 mm, D 2 mm, Fig. 1). Additionally,
the products of the objective function used in the
registration procedure (i.e. cross-correlations, Fig. 2)
were entered in an analysis of variance, as a means of
cross-validation of the voxel-based analysis of variances.
2.1. EEG procedure
While subjects were performing a visual selective
attention task (for a detailed descripition, see Jonkman
et al., 1997), EEG activity was measured from 62 tin
electrodes placed on the scalp by means of an electrocap. A
reference electrode was attached to the left mastoid, and a
ground electrode was placed in the middle of the forehead.
Impedances of ground and reference electrodes were kept
below 5 k O. Horizontal EOG was recorded from electrodes
attached to the outer canthus of each eye. Vertical EOG was
measured from infra- and supra-orbitally placed electrodes
at the left eye. All signals were amplified with a time constant
of 10 s by Sensorium EPA-5 amplifiers (Sensorium, Inc.,
Charlotte, VT), digitized on-line by a computer at 256 Hz and
stored as a continuous signal. After completion of the task,
electrode positions and five fiducial marker positions(nasion, left and right pre-auricular points and mastoids)
were digitized by means of a Polhemus Isotrak digitizer
(Polhemus, Inc., Colchester, VT) for coregistration with
individual MRI scans. Oil-filled capsules were placed on the
digitized marker positions to make them clearly visible in the
MR images.
Signals were epoched off-line starting 100 ms before
stimulus onset and lasting for 1 s. Epochs were filtered with
a 30 Hz, 24 dB/octave digital low pass filter and baseline
corrected on the basis of the 100 ms pre-stimulus interval.
Epochs containing artifacts like flat lines, saturation of the
A/D converter and amplitudes larger than G125 mV were
removed. EOG artifacts were removed from the EEG by
subtracting vertical and horizontal EOG from the EEG
epochs by a regression method in the time domain
(Kenemans et al., 1991). ERPs for unattended standard
and deviant stimuli were computed by averaging all
remaining trials with correct performance for each subject
per lead. For localizations of P1, the ERPs to unattended
stimuli were averaged for each subject.
All source localizations were done with the Curry
program (Neuroscan, Inc., El Paso, TX). Individual realistic
boundary element (BEM) models of the head were
constructed by an automated procedure (Wagner, 1998).
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–11941190
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 4/7
The unregistered (i.e. non-normalized) scans were used for
this procedure. Using the scaled scans for dipole locali-
zations would result in a localization error, since the scalp
potentials are not scaled accordingly. Therefore, we chose to
use the unregistered scans for localization purposes and to
convert the resulting dipole locations to standardized space
afterwards with the individual registration parameters
resulting from the normalization procedure. For each
subject, the digitized electrode cloud was fitted to the
head model through manual identification of the marker
positions visible in the MR images, which corresponded to
the digitized fiducial marker positions.
Localizations of P1 were based on the averaged ERPs for
unattended stimuli in order to have maximal signal to noise
ratios. All leads were re-referenced to a common average
reference, and a mirrored fixed dipole pair was chosen to
model P1. An initial dipole solution was produced based on
the grand average ERPs for each group on a normalized
averaged head model (nZ50). The time point for this
solution was chosen as a 25 ms window around the peak of
Fig. 1. Coronal sections showing areas of significant differences in voxel variance ( P!0.01) in group-wise comparisons.
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–1194 1191
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 5/7
the global field power in the P1 latency (approximately 100–
125 ms). This grand average solution was used to seed the
individual fits. First, a fit was performed with the fixed grand
average dipole locations to compute the explained variance
of this source configuration. For individual fits, the time
point for final reconstructions were chosen similar to the
grand average fit described above, but now the point where
the explained variance was maximal was taken instead of
the maximum global field power (Latency of best fit,
Kenemans et al., 2002). Finally, all source parameters were
optimized. The locations of the dipoles were transformed
with the parameters resulting from the MR normalization
procedure in order to register them into a standardized
coordinate system, enabling between subjects comparisons.
Separate ANOVAs for dipoles were done for each of the
three location parameters ( x, y, z) with Age and Diagnosis as
between factors. Dipole locations in three orthogonal slices
are depicted in Fig. 3.
3. Results
The analysis of objective function values showed
that there was a significant effect of age (F (1,42)Z16.134,
PZ0.000), indicating that the cross-correlations between
target and template brain were significantly lower in the
young groups than in the adolescent groups. There was no
effect of diagnosis.
Overall pair-wise comparisons of whole-volume pooled
variance did not reveal any significant differences between
groups (pooled variance for young PDD 2704.46; young
controls 3451.50; adolescent PDD 2937.57; adolescent
controls 3144.46, ns).The group-wise comparisons of voxel variances pre-
sented in Fig. 1 indicate that the largest differences in the
registration procedure were found in white matter and
thalamic regions in the comparison of the young PDD vs.
young controls. Similar differences were not encountered
when the young groups were compared to adolescent brains.
In all other comparisons, significant differences in variance
seemed to be mostly confined to scalp and neck regions and
to the larger cavities of the head.
The analyses of the dipole locations (Fig. 3) showed
that y-location differed with age, indicating more
Fig. 2. Mean objective function values (cross-correlations) per group. The
young groups show significantly lower cross-correlations than adolescents.
Fig. 3. Orthogonal views of dipole locations of P1 in all groups.
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–11941192
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 6/7
anterior locations in the older age group (left dipole:
F (1,32)Z16.201, PZ0.000, right dipole F Z13.848,
PZ0.001). No further effects were found.
4. Discussion
The results suggest that the automatic linear registration
procedure used in this study (Collins et al., 1994) performs
less adequate in pediatric groups. The data show that the
final objective values in the pediatric groups were
significantly lower than in the adolescent groups. This
difference could be the result of a structural bias or of
random differences in normalization. The overall group
variances did not differ between groups, but the fact that the
young PDD group showed the lowest overall variance,
paired with a relatively low objective value points in the
direction of a structural bias in normalization in this group.
In the young controls, the highest overall variance was
found. Together with the low objective value, this can be
interpreted as this group showing more random differences
in normalization.
In the regional analysis of voxel-variance, differences
became evident in white matter areas and regions around the
thalamus in the comparison of the young control and PDD
groups. However, no differences were found in cortical grey
matter areas. Therefore, we conclude that in the present
samples differences in the quality of normalization do not
result in differences in the quality of the corresponding head
models for EEG source localization.
A likely explanation for the larger registration differ-
ences in young groups is found in the template brain used inthe normalization procedure. The MNI-template is an
average brain constructed from 305 young adult brains,
and thus shows less similarity with the pediatric scans than
with the adult scans in our study.
The differences found in the scalp and neck regions in the
other comparisons are of little consequence for functional
imaging studies. It is clear that the largest deviations are
most likely to be found in the regions where the change of
contrast in the image is largest, e.g. at the borders of the
head and of the skull cavities. In brains where the linear
registration is less successful, variances will be large at
these borders as opposed to, e.g. white matter areas where
the distribution of grey-values is more homegeneous. In
fMRI analyses, the skull will likely be removed from the
image since it is unnecessary for the functional analysis. For
electrophysiological source localisation studies, the skull is
needed for the construction of a model of the head. We
chose to build the head model and to do the source analyses
on the unregistered scans, and to align the resulting sources
with the registered volume afterwards. If the source
reconstructions are done on scans of which the physical
dimensions have changed due to the registration procedure,
it is necessary to scale the scalp-recorded raw data as well.
Using unregistered scans for the actual source modeling
solves this problem and also does not introduce the
registration errors of the skull region in the head model.
Still, the resulting source locations have to be registered to
standardized space in order to facilitate group comparisons
and statistical analyses, and thus proper spatial normal-
ization remains of pivotal importance.
Voxelwise F -tests were used to test for differencesbetween group variances. Concluding from the results that
there are no such differences in variances incorporates the
problem of confirming the null hypothesis. However, in the
current comparisons voxels with a significance level of 1%
and lower were marked. This is very liberal; considering
that for each contrast 2563 comparisons were made, errors in
registration had every chance to show up throughout the
entire volume. It is important to note that the head only
comprises about 30% of the entire volume. Given the
statistical threshold of 1% used in this study, averaged over
volumes 3.97% of all voxels was labeled as significantly
different. On average 81% of these significant voxels were
located in the head. Given the large variation in grey values
within the head, we would expect the majority of significant
voxels in this area. Furthermore, 3.97% voxels identified as
significant is indeed more than would be expected by
chance, based on a 1% threshold. Given the figures noted
above, there is almost a 10% positive rate within the head,
yet significant differences were only encountered in certain
head regions and not in gyral patterns or throughout the
brain.
The larger variance differences and lower cross-corre-
lation values found in the present study may indicate that the
registration of pediatric data was less successful, and that
one should be cautious especially when using the procedurein comparisons of pediatric images. Also, dipole analyses of
P1 did not reveal an effect of diagnosis on dipole location,
but dipoles in the older age group were located more
anteriorly. This effect may be the result of further
maturation of the visual areas (Allison et al., 1984; Dustman
et al., 1982; Ossenblok et al., 1994) and of larger
contributions of frontocentral areas to the scalp distribution
(Dustman et al., 1982). To what extent the different spatial
normalization in the pediatric groups has contributed to the
age effect remains inconclusive. Differences in spatial
normalization may be a confound in the interpretation of
source localizations, and the present paper provides a simple
method to check for such differences. A limitation of the use
of the cross correlations to assess whether the individual
brains were adequately matched to the template is that this
same measure was used as a criterion in the normalization
procedure. Therefore, it should be supplemented with
additional measures of normalization accuracy and var-
iance, like the variance measures as presently proposed. In
turn, a possible complication with the use of such variance
measures may lie in the way they depend on aspects of data
processing, most notably global normalization of image
intensities. For example, although the presently applied
intesity normalization serves in general to increase
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–1194 1193
8/2/2019 Hoeksma Et Al Clin Neurophys (2005) 116(5) 1188-1194
http://slidepdf.com/reader/full/hoeksma-et-al-clin-neurophys-2005-1165-1188-1194 7/7
the comparability among individual datasets, its precise
effects on variance measures should be further investigated.
In addition, a systematic bias in normalization to a template
is not revealed in voxelwise group variance, and therefore
the latter measure should be supplemented by other
measures like the objective function.
References
Allison T, Hume AL, Wood CC, Goff WR. Developmental and aging
changes in somatosensory, auditory and visual evoked potentials.
Electroencephalogr Clin Neurophysiol 1984;58:14–24.
Carper R, Moses P, Tigue Z, Courchesne E. Cerebral lobes in autism: early
hyperplasia and abnormal age effects. Neuroimage 2002;16:1038–51.
Collins DL, Neelin P, Peters TM, Evans AC. Automatic 3D intersubject
registration of MR volumetric data in standardized Talairach space.
J Comput Assist Tomogr 1994;18:192–205.
Courchesne E, Press GA, Yeung-Courchesne R. Parietal lobe abnormalities
detected with MR in patients with infantile autism. AJR Am J
Roentgenol 1993;160:387–93.
Courchesne E, Karns CM, Davis HR, Ziccardi R, Carper RA, Tigue ZD,
Chisum HJ, Moses P, Pierce K, Lord C, Lincoln AJ, Pizzo S,
Schreibman L, Haas RH, Akshoomoff NA, Courchesne RY. Unusual
brain growth patterns in early life in patients with autistic disorder: an
MRI study. Neurology 2001;57:245–54.
Dustman RE, Shearer DE, Snyder EW. Age differences in augmentin-
g/reducing of occipital visually evoked potentials. Electroencephalogr
Clin Neurophysiol 1982;54:99–110.
Evans AC, Collins DL, Milner B. An MRI-based stereotactic brain atlas
from 300 young normal subjects, 408, Anaheim. In: Proceedings of the
22nd symposium of the society for neuroscience.
Fuchs M, Wagner M, Kastner J. Boundary element method volume
conductor models for EEG source reconstruction. Clin Neurophysiol
2001;112:1400–7.
Jonkman LM, Kemner C, Verbaten MN, Koelega HS, Camfferman G, vdGaag RJ, Buitelaar JK, Engeland H. Effects of methylphenidate on
event-related potentials and performance of attention-deficit hyper-
activity disorder children in auditory and visual selective attention
tasks. Biol Psychiatry 1997;41:690–702.
Kenemans JL, Molenaar PC, Verbaten MN, Slangen JL. Removal of the
ocular artifact from the EEG: a comparison of time and frequency
domain methods with simulated and real data. Psychophysiology 1991;
28:114–21.
Kenemans JL, Lijffijt M, Camfferman G, Verbaten MN. Split-second
sequential selective activation in human secondary visual cortex. J
Cogn Neursci 2002;14:48–61.
Levine BK, Beason-Held LL, Purpura KP, Aronchick DM, Optican LM,
Alexander GE, Horwitz B, Rapoport SI, Schapiro MB. Age-related
differences in visual perception: a PET study. Neurobiol Aging 2000;21:577–84.
Levy KL. An empirical comparison of several multiple range tests for
variances. J Am Stat Assoc 1975;70:180–3.
Lord C, Rutter M, Le Couteur A. Autism diagnostic interview—revised: a
revised version of a diagnostic interview for caregivers of individuals
with possible pervasive developmental disorders. J Autism Dev Disord
1994;24:659–85.
Maintz JBA, Viergever MA. A survey of medical image registration. Med
Image Anal 1998;2:1–36.
Onofrj M, Thomas A, Iacono D, D’Andreamatteo G, Paci C. Age-
related changes of evoked potentials. Neurophysiol Clin 2001;31:
83–103.
Ossenblok P, De Munck JC, Wieringa HJ, Reits D, Spekreijse H.
Hemispheric asymmetry in the maturation of the extrastriate checker-
board onset evoked potential. Vis Res 1994;34:581–90.Piven J, Berthier ML, Starkstein SE, Nehme E, Pearlson G, Folstein S.
Magnetic resonance imaging evidence for a defect of cerebral cortical
development in autism. Am J Psychiatry 1990;147:734–9.
Piven J, Arndt S, Bailey J, Andreasen N. Regional brain enlargement in
autism: a magnetic resonance imaging study. J Am Acad Child Adolesc
Psychiatry 1996;35:530–6.
Scherg M. Fundamentals of dipole source potential analysis. In: Grandori F,
Hoke M, Romani GL, editors. Auditory evoked magnetic fields and
electric potentials. Basel: Karger; 1990. p. 40–69.
Sled JG, Zijdenbos AP, Evans AC. A nonparametric method for automatic
correction of intensity nonuniformity in MRI data. IEEE Trans Med
Imaging 1998;17:87–97.
Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain: 3-
Dimensional proportional system—an approach to cerebral imaging.
New York: Thieme Medical Publishers; 1988.
Toga AW, Thompson PM. The role of image registration in brain mapping.
Image and vision computing 2001;19:3–24.
Wagner M. Rekonstruktion neuronaler Strome aus bioelektrischen und
biomagnetischen Messungen auf der aus MR-Bildern segmentierten
Hirnrinde. Aachen: Shaker Verlag; 1998.
M.R. Hoeksma et al. / Clinical Neurophysiology 116 (2005) 1188–11941194