magnetoencephalography is feasible for infant assessment...
TRANSCRIPT
www.elsevier.com/locate/yexnr
Experimental Neurology
Magnetoencephalography is feasible for infant assessment of
auditory discrimination
Marie Cheoura,b,c,*, Toshiaki Imadad,e, Samu Tauluf, Antti Ahonenf,
Johanna Salonena,c, Patricia Kuhld
aDepartment of Psychology, University of Miami, Coral Gables, FL 33124-0751, USAbDepartment of Pediatrics, University of Miami School of Medicine, Miami, FL 33124, USA
cBioMag Laboratory, Helsinki University Central Hospital, Helsinki, FinlanddInstitute for Learning and Brain Sciences, University of Washington, Seattle, WA 98195, USA
eResearch Center for Advanced Technologies, Tokyo Denki University, Inzai, Chiba, JapanfElekta Neuromag Oy, Helsinki, Finland
Received 10 December 2003; revised 11 June 2004; accepted 30 June 2004
Available online 15 September 2004
Abstract
Magnetoencephalography (MEG) detects the brain’s magnetic fields as generated by neuronal electric currents arising from synaptic ion
flow. It is noninvasive, has excellent temporal resolution, and it can localize neuronal activity with good precision. For these reasons, many
scientists interested in the localization of brain functions have turned to MEG. The technique, however, is not without its drawbacks. Those
reluctant to employ it cite its relative awkwardness among pediatric populations because MEG requires subjects to be fairly still during
experiments. Due to these methodological challenges, infant MEG studies are not commonly pursued. In the present study, MEG was
employed to study auditory discrimination in infants. We had two goals: first, to determine whether reliable results could be obtained from
infants despite their movements; and second, to improve MEG data analysis methods. To get more reliable results from infants we employed
novel hardware (real-time head-position tracking system) and software (signal space separation method, SSS) solutions to better deal with
noise and movement. With these solutions, the location and orientation of the head can be tracked in real time and we were able to reduce
noise and artifacts originating outside the helmet significantly. In the present study, these new methods were used to study the biomagnetic
equivalents of event-related potentials (ERPs) in response to duration changes in harmonic tones in sleeping, healthy, full-term newborns.
Our findings indicate that with the use of these new analysis routines, MEG will prove to be a very useful and more accessible experimental
technique among pediatric populations.
D 2004 Elsevier Inc. All rights reserved.
Keywords: Auditory discrimination; Event-related potentials (ERPs); Infants; Magnetoencephalography (MEG)
Introduction
The investigation of auditory discrimination in infants
has largely been done via behavioral methods (for a review,
see Kuhl, 2000; Stockard-Pope, 2001; Werker and Rubel,
1992). More recently, however, direct brain measurements
0014-4886/$ - see front matter D 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.expneurol.2004.06.030
* Corresponding author. Department of Psychology, University of
Miami, PO Box 249229, Coral Gables, FL 33124-0751. Fax: +1 305 284
3402.
E-mail address: [email protected] (M. Cheour).
such as functional magnetic resonance imaging (fMRI)
(Dehaene-Lambertz and Pena, 2001), optical topography
(Pena et al., 2003), and especially electroencephalography
(EEG) (Cheour et al., 1997; Dehaene-Lambertz et al., 2002;
for a review see Cheour et al., 2001) have been employed in
pediatric populations. Unfortunately, each of these methods
has its limitations, due to, for instance, spatial (EEG) or
temporal resolution (fMRI, optical topography). Moreover,
many other factors besides brain functions affect results
obtained by EEG. Extracellular flow, for example, is
affected significantly by the differing electrical conductivity
190 (2004) S44–S51
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51 S45
properties of the scalp, skull, cerebrospinal fluid, and the
brain itself (Lounasmaa et al., 1996).
Although it has been available for more than 35 years,
having been invented in 1968 (Cohen, 1968), MEG has only
rarely been employed among pediatric populations. MEG is
noninvasive and offers excellent temporal and spatial
resolution, especially when multiple active sources are
compared within a single experimental session. It works
by detecting magnetic fields generated by the brain’s
cellular currents during synchronized neural activity.
Among the few infant MEG experiments is a study by
Paetau et al. (1994) in which data from a single 3-month-old
infant and from a group of older children and adults are
reported. In this study, tones, tone pairs, and pseudowords
were presented to the subjects. The authors reported that all
stimulus types elicited substantially similar responses in all
age groups. However, in the infant, the authors did not see
any N1m response but a large positivity P1m peaking at 190
ms and dominating the response up to 500 ms. Another
example of the rare infant MEG experiments is a study by
Lengle et al. (2001) on twenty 2- to 6-week-old infants.
Authors report that in all neonates tones presented to them
elicited a dominant component that was negative toward the
face and positive toward the occiput. Huotilainen et al.
(2003) also published recently the infant MEG responses
recorded from the infant’s left hemisphere, where they
showed mismatch activity in response to a frequency-
deviated sound.
Recently, a few MEG fetus studies have also been
published. Blum et al. (1985) were the first to demon-
strate that MEG can be used to obtain fetal auditory
evoked responses. Since then, some other researchers
have applied this technique to study fetuses. These studies
have reported that auditory evoked potentials can be
detected in 28–80% of the subjects, depending on, for
example, how many times MEG recordings were
performed for each subject (Eswaran et al., 2002a,b;
Lengle et al., 2001; Schneider et al., 2001; Wakai et al.,
1996; Zappasodi et al., 2001).
Four factors help to explain why infant MEG studies are
still rare. Naturally, the higher price of MEG as compared to
EEG might be one of the reasons why MEG infant studies
are so rare. Second, to obtain reliable results in MEG
experiments, it is important that the subject remains
relatively still. Obviously, this is difficult to achieve in
infants. Third, MEG requires that the infant’s head is close
to the MEG helmet wall to obtain reliable results. This
presents a challenge because existing helmets are designed
for adults, making them too large to offer uncompromised
accuracy in infant studies. Fourth, factors such as the
proximity between the heart and the MEG helmet can
greatly influence the MEG signal. The infant heart is closer
to the head than it is in adults and infant heart beats,
approximately 10 times larger than their brain magnetic
fields at MEG sensor location, largely disturbs MEG
recordings.
In the present study, as the first report of a series of
experiments on neonates, 6-month-olds, and 12-month-olds
using simple tones, complex harmonic tones, and syllables,
we report preliminary results indicating the feasibility of
using MEG to study auditory discrimination in neonates.
Our findings rely on the invention of MEG techniques that
helped us to overcome some of the aforementioned
problems with infants. First, we localized the head position
and orientation during each epoch, making it possible to
average the epochs by selecting those in which the head
position was close enough to a predefined reference
position. Secondly, we employed an algorithm, called the
signal space separation (SSS) method, that greatly reduces
magnetic disturbances that originate outside the multi-
channel sensor helmet by separating the signal space that
produces brain signals (those inside the sensor helmet) from
the disturbance space outside the sensor helmet (Taulu et al.,
in press). The magnetic field distributions over the sensor
volume that are induced by currents from the inside are
fundamentally different from those induced by currents on
the outside, allowing us to effectively reduce magnetic noise
originating from outside the helmet. The combination of
these features made it possible to use the magnetic
responses of infants to examine their abilities to discriminate
auditory signals.
Materials and methods
Subjects
Eight full-term newborns (five females, gestational age
[GA]: 37–42 weeks) participated in this study. Parents
provided written informed consent for their infants to
participate in the experiments. Infants were tested 1–5 days
after birth. During the recordings infants were asleep.
Stimuli
An oddball sequence was employed, including a single
standard (P = 85%) and a single deviant stimulus (P =
15%) that differed in duration; the standard sounds were 100
ms in duration and the deviant sounds were 40 ms. All the
sounds were composed of three sinusoidal partials of three
500-Hz harmonic sounds, that is, 500, 1000, and 1500 Hz.
The constant stimulus-onset asynchrony was 800 ms.
Stimulus sequences were pseudo-random and each deviant
tone was preceded by at least two standard tones. The
stimuli were delivered via nonmagnetic headphones to the
right ear of each subject. The duration of the experiment was
about half an hour.
Recording
All experiments were conducted in a magnetically
shielded room of the BioMag Laboratory of the Helsinki
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51S46
University Central Hospital. Although the helmet geometry
of whole-head 306-channel MEG system (Elekta Neuro-
magR, Elekta Neuromag Oy, Helsinki, Finland) is designed
for adult measurements, this system was employed to record
the infant’s auditory magnetic responses from their left
hemisphere. In the Elekta NeuromagR system, the magnetic
field is sampled with 510 superconducting sensor loops at
distinct locations. One hundred two loops are used as
magnetometers and the remaining 408 are hardwired to
form 204 planar gradiometers. The bases of the gradiom-
eters are orthogonal and their midpoints coincide. First, the
planar gradiometers measure two orthogonal spatial deriv-
atives of Bz (BBz/Bx and BBz/By); Bz is the radial
component of the brain’s magnetic field B. For idealized
dipolar sources, this planar-type gradiometer shows the
maximum amplitude right above a local current source in
the brain. The baseline length of each planar gradiometer is
16.5 mm and the size of this pickup coil is 2 � 10 � 28 mm.
Second, a single magnetometer records Bz. As a result, the
spatial distributions of Bz, BBz/Bx, and BBz/By are recorded.
Average physical center-to-center distance between sensors
is 35.7 F 1.1 mm, which corresponds to effective channel
separation at 20 mm (see Ahonen et al., 1993). The origin of
the device coordinate system is at the center of the sensor
helmet with the x-axis directing from left to right channel
and the y-axis from posterior to anterior channel. The left
side of each infant’s head was positioned over a part of the
306-channel sensor array that would correspond to the adult
occipital region.
Epochs of 800-ms duration (including a 100-ms presti-
mulus baseline) were digitized at 600 Hz (analog filter pass
band 0.1–172 Hz). During the recording, each subject’s
head location and orientation with respect to the device’s
coordinate system was recorded (every 167 ms; carrier
frequencies: 154, 158, 162, and 166 Hz) by the continuous
head monitoring system.
Analysis
Because we recorded the head position and orientation
every 167 ms, first we calculated the head position and
orientation of every epoch. Epochs that had the head
location closer than 5 mm to the median head location of
all recorded epochs were accepted to be averaged. There-
fore, epochs that had the head position more than 5-mm far
away from the median location were rejected. Subsequently,
artifacts arising from outside the sensor array, such as those
stemming from infant heart beat, limb movement, or other
ambient magnetic disturbances, were greatly reduced by the
signal space separation method (SSS) (Taulu et al., in press).
This method efficiently separates brain signals from external
disturbances based on fundamental properties of the
magnetic fields.
The SSS method is based on the fact that a modern MEG
device, comprised of more than 300 signal channels,
provides generous spatial oversampling of both biomagnetic
and external disturbance magnetic fields. This is true for all
sources located more than a couple of centimeters from the
nearest sensor in the array, an assumption valid for the
possible magnetic sources in the current MEG devices. As
the spatial complexity of the field decreases as a function of
distance, the number of channels exceeds the essential
number of degrees of freedom of the measured field even for
the nearest sources. Consequently, a relatively low-dimen-
sional magnetic subspace spanning all measurable magnetic
fields can be determined.
Because the sensor array is located in a source-free
volume between the volume of interest and the volume
containing all sources of external interference, it turns out
that the magnetic subspace can be split into two separate,
linearly independent subspaces: one containing signals of
interest to us coming from the volume surrounded by the
sensors and another containing signals from sources outside
the array. As the former volume contains the biomagnetic
signal sources and the latter volume contains the external
disturbance sources, any measured signal can be uniquely
decomposed into the magnetic subspace with separate
coefficients corresponding to the subspace spanning the
biomagnetic signals and to the subspace spanning the
external disturbance signals.
The basic assumption of having a source-free sensor
volume means that the magnetic field b can be expressed as
a gradient of a harmonic scalar potential V in that volume.
The harmonicity of the potential allows one to express V as
an expansion of harmonic basis functions, for example,
spherical harmonic functions. Furthermore, this expansion
can be separated into two different expansions; one for the
sources closer to the center of the expansion than any of the
sensors and another for sources more distant to the center of
the expansion than any of the sensors. Consequently,
different basis signal vectors can be determined for signals
produced by internal and external sources. In this way, the
linearly independent SSS basis is formed for modern
multichannel measurement devices with a sufficiently high
number of channels allowing one to elegantly reduce the
external interference signals.
After completing the SSS process, the averaged data
were digitally lowpass filtered (cutoff, 20 Hz; decay, 48 dB/
oct), and then the linear trend was eliminated by removing
the DC components of the 100-ms prestimulus and the 100-
ms tail baselines; the 100-ms tail baseline is equivalent to
the 100-ms prestimulus baseline of the next epoch.
Because the head location and orientation determined by
the median location during each subject’s measurement
were not equivalent across subjects, each subject’s original
head coordinate system was translated and rotated to
correspond to a reference or standard head location; the
head coordinate system has its x-axis directing from the left
to right pre-auricular point and its y-axis from the origin to
the nasion. To convert into the standard head location, each
subject’s head was virtually moved as follows. First, the
center of a sphere approximated to each subject’s brain was
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51 S47
moved to the point (0.0 �30.0 10.0) [mm] of the device
coordinate system. Second, each subject’s head was rotated
so that all three axes of its head coordinate system ran
parallel to the axes of the device coordinate system (see Fig.
1). We assumed that the center of the approximated sphere
of the newborn infant’s brain was at (0.0 4.7 36.6) [mm]
with respect to the head coordinate system and its radius
was 50.0 [mm]; these specific values were obtained by
reducing the adults’ average brain size and approximated
sphere. The magnetic data at each channel were converted
using minimum norm estimates (Uutela et al., 2001) as if the
infant’s head position had been at the standard head position
and orientation.
The difference waveforms were obtained by subtracting
the averaged response to repetitively presented standard
stimuli from the response to occasionally presented deviant
stimuli. After the head position standardization, the follow-
ing two analyses were performed.
The magnetic amplitude of the gradiometer was calcu-
lated at 10 channel locations that have distances less than 65
Fig. 1. Effect of SSS algorithm. In the left panel, responses of subject SH to t
hemisphere after lowpass filter (cutoff: 40 Hz) processing and a baseline (�100 to
described in the inset. Waveforms in triple sensor 25 are enlarged on the right pan
waveforms after applying the SSS algorithm. In the right panel, the left upper fram
lower frame shows the waveforms recorded by an x-axis-oriented gradiometer, an
directions of x-axis and y-axis, shown in the inset, are based on the channel coordi
the channel location and direction on the sensor helmet. A filled green circle ind
leftmost point of the approximated sphere, in the original measurement location. A
head position standardization; in Figs. 2 and 3, waveforms, contour maps, and
Materials and methods section.
mm to the leftmost point of approximated sphere, which
was (�50.0 4.7 36.6) [mm] with respect to the head
coordinates (see Fig. 2). Because we do not have each
subject’s magnetic resonance images, this leftmost point is
supposed, based on adult brain structure, to be located at the
leftmost point of the left temporal region along or close to
the Sylvian fissure, which is the auditory region. To obtain a
good S/N ratio comparable to the adult measurement, the
reference distance 65 mm was determined in such a way that
95% of the Elekta NeuromagR channels are closer than this
distance to the approximated sphere when the adult sits in
the average position (calculated from 206 experiments). If
the distance is longer, the signal from the brain is very small,
resulting in worse S/N ratio. The waveforms from the
gradiometers are analyzed because they show a maximum
peak right above the neuronal current (good spatial
correspondence) and also show an approximate direction
of the neuronal current beneath that channel. We mainly
analyzed the difference waveforms. To find the peak
amplitudes and latencies of the difference waveforms, we
he deviant stimulus are shown for 18 channel locations covering the left
0 ms) correction. The location of brain relative to the channel locations is
el, where thin red curves are original waveforms and thick blue curves are
e shows the waveforms recorded by a y-axis-oriented gradiometer, the left
d the right frame shows the waveforms recorded by a magnetometer. The
nate system, and therefore the absolute directions are different depending on
icates the channel closest to a point (�50.0 4.7 36.6) [mm], which is the
n open green circle shows the location closest to the above point after the
arrow maps are always illustrated after head position standardization. See
Fig. 2. Waveforms from individuals. Difference waveforms of the four individual subjects. Channels in the occipital area of an adult-size sensor helmet were
used in the actual recording of magnetic responses from the neonates’ left hemisphere, but the waveforms here were shown after the head position
standardization. The channels, numbered from 1 to 10, are 10 channel locations closest to the leftmost point of the approximated sphere (�50.0 4.7 36.6) [mm]
in a shorter distance order.
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51S48
divided the latency range into three sections based on visual
inspection of the data and previous studies (compare Cheour
et al., 1998); the first range from 40 to 100 ms correspond-
ing to event-related potential’s (ERP’s) early component
(EC), the second one from 200 to 300 ms corresponding to
ERP’s mismatch negativity (MMN), and the third one from
350 to 550 ms corresponding to ERP’s late discriminative
negativity (LDN). We identified the peaks separately
depending on the current orientation in terms of ERP
measurement, namely positive and negative current, where
the negative current flows through the auditory region from
the superior to inferior direction, resulting in a vertex
negativity in an EEG recording. For further analysis, we
employed the peak whose amplitude has the S/N of more
than 2. The noise level was the standard deviation of the
first baseline (�100 to 0 ms). The amplitude refers to the
vector sum of a pair of orthogonal gradiometer outputs.
Results
Using the real-time head-position tracking system, we
could effectively reject the epochs in which the head
position was farther than 5 mm from the reference head
position. The SSS algorithm reduces noise components from
sources originating outside the sensor helmet (Taulu et al.,
in press). Fig. 1 illustrates the effects on gradio- and
magnetometer channels in a representative infant. The inset
of Fig. 1 shows a sketch of the brain to show its
approximate location relative to the sensor channels after
head position standardization.
After the SSS process, large artifacts still remained in
four out of eight subjects. We regarded that the data were
contaminated by the large artifact when the maximum
waveform peak, calculated from the waveforms digitally
filtered at 20 Hz and with removal of the DC component
during the two baselines, resides outside the latency range
from 60 to 700 ms. These data were eliminated from further
analysis, resulting in a final analysis of MEG data from four
females (age: 2–5 days, GA: 38–42 weeks). Fig. 2 shows
the difference waveforms obtained from four subjects at 10
triple sensor locations (20 gradiometers and 10 magneto-
meters), covering most of the left hemisphere. Large
deflections are seen mainly in the channels close to the left
auditory region.
As indicated in the Materials and methods section, the
difference waveforms at the 10 triple sensor locations shown
in Fig. 2 were analyzed in two ways after head position
standardization. Table 1 shows the significant negative and
positive peak data for the three latency ranges. In the first
latency range (40–100 ms), one subject showed a negative
peak and two subjects showed a positive peak. In the second
latency range (200–300 ms), one subject showed a negative
peak and three subjects showed a positive peak. In the third
Table 1
Individual peak data
Subject Negative peak Positive peak
Latencya Amplitudeb S/N Sensor Latencya Amplitudeb S/N Senso
(a) Latency range: 40–100 ms
ah 56.61 45.95 2.93 6
sh
ch 84.91 60.75 4.27 2
bh 61.6 30.16 4.85 6
(b) Latency range: 200–300 ms
ah 276.38 86.42 5.51 6
sh 234.76 35.97 10.19 7
ch 259.73 80.74 5.15 3
bh 229.76 53.21 8.55 6
(c) Latency range: 350–550 ms
ah 382.94 110.64 7.06 6
sh 431.22 118.1 5.35 9
ch 497.82 105.7 6.74 3
bh 454.53 39.04 9.07 3
Each subject’s waveform peak amplitude, latency, S/N (amplitude to baseline noise) ratio, and the sensors that show those peaks are shown separately in each
of three latency ranges: (a) 40–100 ms, (b) 200–300 ms, and (c) 350–550 ms. bNegative peakQ corresponds to the peak that shows a negative current from
superior to inferior region. bPositive peakQ refers to the peak that has a positive current from inferior to superior region.a ms.b fT/cm.
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51 S49
latency range (350–550 ms), two subjects showed a
negative peak and two subjects showed a positive peak.
Except for one case in the first latency range, all significant
peaks have the S/N of more than 4.
Fig. 3 shows each individual’s contour maps and arrow
maps at the maximum peak latencies (Table 1) in each of
three latency ranges. The arrow maps show the approximate
current flow (direction and relative magnitude) beneath each
channel location.
Discussion
Our results demonstrate that MEG is a promising tool for
recording auditory ERP in infants. The real-time head-
position tracking system enabled us to average the epochs
by rejecting the epochs in which head location deviated
significantly from the reference head position. Moreover,
the new algorithm reduced the magnetic interference from
outside the sensor helmet in our newborn baby recording as
well.
The waveform peak analysis showed that at the first
latency range (40–100 ms) there was no clear consistent
tendency across subjects but one subject showed a negative
peak and two subjects showed a positive peak. The second
latency range (200–300ms) showed a trend of a positive peak
although one subject showed a negative peak. The third
latency range (350–550 ms) did not show a consistent
tendency across subjects either because two subjects showed
a negative peak and two subjects showed a positive peak.
Although we could not find a consistent magnetic
counterpart of ERP’s early component, two subjects showed
r
a magnetic peak corresponding to the positive current. In
some previous EEG studies, early negativity has been
reported instead of positivity. This has typically not been
very large in amplitude and, according to some studies,
seems to appear so quickly after the stimulus presentation
that it may include some middle-latency components
(Cheour et al., 1997; Ceponiene et al., 2002; Pihko et al.,
1999). If this is the case, it is unlikely that MEG would
detect these deep sources. One also needs to take into
account that due to individual variation in ERP latencies in
infants, it is likely that in the figures presenting group data
this EC often seems to be overlapped by the next
component, MMN (Ceponiene et al., 2002).
The magnetic pattern of a positive current continued from
the first latency range to the second latency range; thus, from
40 to 300 ms. In the second latency range (200–300 ms), a
significant peak was detected in all subjects. In three out of
four subjects, it showed a magnetic deflection corresponding
to a positive current and in one subject it showed a magnetic
deflection corresponding to a negative current.
Previous infant EEG findings in this latency range have
been contradictory. Some studies have reported only
negativities (Cheour et al., 1997; 2002a,b; 2003; Tanaka et
al., 2001) and some only positive responses (Dehaene-
Lambertz et al., 2002; Pihko et al., 1999; Winkler et al.,
2003). The majority of them, however, have reported both
positive and negative responses (Ceponiene et al., 2002;
Kushnerenko et al., 2002; Kurtzberg et al., 1995; Leppanen
et al., 1997; Morr et al., 2002).
In the third latency range (350–550 ms), a significant
peak was detected in all subjects; two out of four subjects
showed a magnetic deflection corresponding to a positive
Fig. 3. Contour maps and arrow maps from each individual. Contour maps and corresponding arrow maps at the peak latencies in each latency range. In the
contour maps, the magnetic flux comes out of the head in the red line regions and goes into the head in the blue line regions. In the arrow maps, the arrow
direction shows the current direction below each sensor and the arrow size relatively corresponds to the current magnitude. The arrow size is consistent only
within a subject.
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51S50
current and the others showed a magnetic deflection
corresponding to a negative current. Most EEG studies
have reported a large amplitude negative response called
late discriminative negativity (LDN) at this latency range
(Ceponiene et al., 2002; Korpilahti et al., 1995; Martynova
et al., 2003).
Our results show that individual variations seem to be
quite large in infant MEG data. This finding is consistent
with infant EEG studies. Several explanations have been put
forth. Some authors have suggested that infant hearing is
quite immature and it is likely that not all infants are able to
accurately discriminate sounds. This may very well be the
case, especially if simple tones are used, because recent
studies have shown that the infant auditory cortex is, indeed,
quite immature (Moore, 2002).
It has been further proposed that some of the components
may overlap. It has been debated whether large positive
responses commonly reported at the latency of 200–350 ms
reflect a drift in attention, and whether they obscure an
underlying negative component (Ceponiene et al., 2002;
Kushnerenko et al., 2002). Although it seems unlikely that
this positive response solely reflects orientation toward the
stimuli and therefore is an infant version of the P3a
component, it seems nonetheless likely that there are two
different components appearing in infant ERPs at the
latency range of 200–300 ms. Unfortunately, this debate
has thus far been futile because EEG offers only limited
capacity to separate two different components. MEG has the
potential to offer, for the first time, the genuine possibility of
separating infant ERP components that have different scalp
distributions.
The present study demonstrates a substantial methodo-
logical improvement in MEG data collection and data
analysis. Disturbances due to the heart and limbs remain a
problem in some infants. In addition, adult-sized helmets are
too large for infants; to improve data quality, infant-sized
helmets should be designed. In spite of these limitations,
however, the present study demonstrates that MEG is a
viable tool for research on infant auditory processing. As
such, MEG will be a valuable asset for future investigations
of infants’ perception of complex auditory stimuli such as
speech.
M. Cheour et al. / Experimental Neurology 190 (2004) S44–S51 S51
Acknowledgments
The authors would like to thank Jussi Nurminen for his
assistance in data collection and Dr. Irving Smith for his
comments on the manuscript. We would also like to thank
the BioMag laboratory of the Helsinki University Central
Hospital for allowing us to collect the data in their
laboratory. Funding for the research was provided by grants
to PKK (NIH HD37954), the Human Frontiers Science
Program (HF159), and the William P. and Ruth Gerberding
University Professorship. Additional support provided by
the Talaris Research Institute, the family Foundation of
Bruce and Jolene McCaw, and the Institute for Learning and
Brain Sciences is gratefully acknowledged.
References
Ahonen, A., Hamalainen, M., Ilmoniemi, R., Kajola, M., Knuutila, J.,
Simola, J., Vilkman, V., 1993. Sampling theory for neuromagnetic
detector arrays. IEEE Trans. Biomed. Eng. 40, 859–869.
Blum, T., Saling, E., Bauer, R., 1985. First magnetoencephalographic
recordings of the brain activity of a human fetus. Br. J. Obstet.
Gynaecol. 92, 1224–1229.
Ceponiene, R., Kushnerenko, E., Fellman, V., Renlund, M., Suominen, K.,
Naatanen, R., 2002. Event-related potential features indexing central
auditory discrimination by newborns. Brain Res. Cogn. Brain Res. 3,
101–113.
Cheour, M., Alho, K., Sainio, K., Reinikainen, K., Renlund, K., Aaltonen,
O., Eerola, O., N77t7nen, R., 1997. The Mismatch negativity to changes
in speech sounds at the age of three months. Dev. Neuropsychol. 13,
167–174.
Cheour, M., Alho, K., Ceponiene, R., Reinikainen, K., Sainio, K.,
Pohjavuori, M., Aaltonen, O., N77t7nen, R., 1998. Maturation of
mismatch negativity in infants. Int. J. Psychophysiol. 29, 217–226.
Cheour, M., Korpilahti, P., Martynova, O., Lang, A., 2001. Mismatch
negativity (MMN) and late discriminative negativity (LDN) in
investigating speech perception and learning in children and infants
[review]. Audiol. Neuro-Otol. 6, 2–11.
Cheour, M., Kushnerenko, E., Ceponiene, R., Fellman, V., N77t7nen, R.,2002a. Electric brain responses obtained from newborn infants to
changes in duration in complex harmonics tones. Dev. Neuropsychol.
22, 471–479.
Cheour, M., Martynova, O., N77t7nen, R., Erkkola, R., Sillanp77, M., Kero,
P., Raz, A., Kaipio, M.-L., Hiltunen, J., Aaltonen, O., Savela, J.,
H7m7l7inen, H., 2002b. Sleep can be used for on-line training of speechsound discrimination in human newborns. Nature 7, 599–600.
Cohen, D., 1968. Magnetoencephalography: evidence of magnetic field
produced by alpha-rhythm currents. Science 161, 784–786.
Dehaene-Lambertz, G., Pena, M., 2001. Electrophysiological evidence
for automatic phonetic processing in neonates. NeuroReport 12,
3155–3158.
Dehaene-Lambertz, G., Dehaene, S., Hertz-Pannier, L., 2002. Functional
neuroimaging of speech perception in infants. Science 298, 2013–2015.
Eswaran, H., Preissl, H., Wilson, J.D., Murphy, P., Robinson, S.E., Rose,
D., Vrba, J., Lowery, C.L., 2002a. Short-term serial magnetoencepha-
lography recordings of fetal auditory evoked responses. Neurosci. Lett.
11, 331128–331132.
Eswaran, H., Wilson, J., Preissl, H., Robinson, S., Vrba, J., Murphy, P.,
Rose, D., Lowery, C., 2002b. Magnetoencephalographic recordings
of visual evoked brain activity in the human fetus. Lancet 7 (360),
779–780.
Huotilainen, M., Kujala, A., Hotakainen, M., Shestakova, A., Kushnerenko,
E., Parkkonen, L., Fellman, V., N77t7nen, R., 2003. Auditory magnetic
responses of healthy newborns. NeuroReport 14, 1871–1875.
Korpilahti, P., Salmela, S., Lang, H., Pfrn, B., Krause, C., 1995. Event-related potentials elicited by complex tones, word and pseudo-words in
normal and language impaired children. Electroencephalogr. Clin.
Neurophysiol. 103, 64.
Kuhl, P.K., 2000. A new view of language acquisition. Proc. Natl. Acad.
Sci. 97, 11850–11857.
Kurtzberg, D., Vaughan, H.G.J., Kreuzer, J.A., Fliegler, K.Z., 1995.
Developmental studies and clinical application of mismatch negativity:
problems and prospects. Ear Hear. 16, 105–117.
Kushnerenko, E., Ceponiene, R., Balan, P., Fellman, V., N77t7nen, R.,2002. Maturation of the auditory change detection response in infants: a
longitudinal ERP study. NeuroReport 28 (13), 1843–1848.
Lengle, J.M., Chen, M., Wakai, R.T., 2001. Improved neuromagnetic
detection of fetal and neonatal auditory evoked responses. Clin.
Neurophysiol. 112, 785–792.
Lepp7nen, P., Eklund, K., Lyytinen, H., 1997. Event-related brain potentialsto change in rapidly presented acoustic stimuli in newborns. Dev.
Neuropsychol. 13, 175–204.
Lounasmaa, O.V., Hamalainen, M., Hari, R., Salmelin, R., 1996.
Information processing in the human brain: magnetoencephalographic
approach. Proc. Natl. Acad. Sci. 20 (93), 8809–8815.
Martynova, O., Kirjavainen, J., Cheour, M., 2003. Mismatch negativity and
late discriminative negativity in sleeping human newborns. Neurosci.
Lett. 10 (340), 75–78.
Moore, J.K., 2002. Maturation of human auditory cortex: implications for
speech perception. Ann. Otol., Rhinol., Laryngol. (Suppl. 189), 7–10.
Morr, M.L., Shafer, V.L., Kreuzer, J.A., Kurtzberg, D., 2002. Maturation of
mismatch negativity in typically developing infants and preschool
children. Ear Hear. 23, 118–136.
Paetau, R., Ahonen, A., Salonen, O., Sams, M., 1994. Auditory evoked
magnetic fields to tones and pseudowords in healthy children and
adults. Helsinki University of Technology Low Temperature Labo-
ratory. Report TKK-F-A722.
Pena, M., Maki, A., Kovacic, D., Dehaene-Lambertz, G., Koizumi, H.,
Bouquet, F., Mehler, J., 2003. Sounds and silence: an optical
topography study of language recognition at birth. Proc. Natl. Acad.
Sci. 30 (100), 11702–11705.
Pihko, E., Leppanen, P.H., Eklund, K.M., Cheour, M., Guttorm, T.K.,
Lyytinen, H., 1999. Cortical responses of infants with and without a
genetic risk for dyslexia: I. Age effects. NeuroReport 6 (10), 901–905.
Schneider, U., Schleussner, E., Haueisen, J., Nowak, H., Seewald, H.J.,
2001. Signal analysis of auditory evoked cortical fields in fetal
magnetoencephalography. Brain Topogr. 14, 69–80.
Stockard-Pope, J.E., 2001. Auditory development and hearing evaluation in
children. Adv. Pediatr. 48, 273–299.
Tanaka, M., Okubo, O., Fuchigami, T., Harada, K., 2001. A study of
mismatch negativity in newborns. Pediatr. Int. 4 (3), 281–286.
Taulu, S., Kajola, M., Simola, J., 2004. The Signal Space Separation
method. Biomed. Tech. 48 (in press).
Uutela, K., Taulu, S., H7m7l7inen, M., 2001. Detecting and correcting for
head movements in neuromagnetic measurements. NeuroImage 14,
1424–1431.
Wakai, R.T., Leuthold, A.C., Martin, C.B., 1996. Fetal auditory evoked
responses detected by magnetoencephalography. Am. J. Obstet.
Gynecol. 74, 1484–1486.
Werker, L.A., Rubel, E.W. (Eds.), 1992. Developmental Psychoacoustics.
American Psychological Association, Washington, DC.
Winkler, I., Kushnerenko, E., Horv7th, J., Ceponiene, R., Fellman, V.,
Huotilainen, M., N77t7nen, R., Sussman, E., 2003. Newborn infants
can organize the auditory world. Proc. Natl. Acad. Sci. 100,
11812–11815.
Zappasodi, F., Tecchio, F., Pizzella, V., Cassetta, E., Romano, G.V., Filligoi,
G., Rossini, P.M., 2001. Detection of fetal auditory evoked responses by
means of magnetoencephalography. Brain Res. 2 (917), 167–173.