www.elsevier.com/locate/ynimg

NeuroImage 28 (2005) 140 – 153

Localization of human supratemporal auditory areas from

intracerebral auditory evoked potentials using distributed

source models

Blaise Yvert,a,* Catherine Fischer,a,b Olivier Bertrand,a and Jacques Perniera

aInserm Unite 280, F-69675, Bron Cedex, FrancebNeurological Hospital, 59 Boulevard Pinel, F-69003 Lyon, France

Received 15 December 2004; revised 21 March 2005; accepted 20 May 2005

Available online 21 July 2005

While source localization methods are increasingly developed to

identify brain areas underlying scalp electro/magnetoencephalographic

data (EEG/MEG), these methods have not yet been used to identify the

sources of intracerebral signals which offer highly detailed information.

Here, we adapted the minimum current estimates method to intra-

cranial data in order to localize supratemporal sources of intracerebral

auditory 1-kHz-tone-evoked potentials occurring within 100 ms after

stimulus onset. After an evaluation of localization method and despite

inter-subject variability, we found a common spatiotemporal pattern of

activities, which involved the first Heschl’s gyrus (H1) and sulcus (HS),

the Planum Temporale (PT), H2/H3 when present, and the superior

temporal gyrus (STG). Four time periods of activity were distinguished,

corresponding to the time range of the scalp components P0, Na, Pa/Pb,

and N100. The sources of the earliest components P0 (16–19 ms) and

Na (20–25 ms) could be identified in the postero-medial portion of HS

or H1. Then, several areas became simultaneously active after 25 ms.

The Pa/Pb time range (30–50 ms) was characterized by a medio-lateral

and postero-anterior propagation of activity over the supratemporal

plane involving successively H1/HS, the Planum Temporale, H2/H3

when present, and the STG. Finally, we found to a large extent that the

N100 (55–100 ms) involved almost the same areas as those active

during the Pa/Pb complex, with a similar propagation of activities.

Reconstructing scalp data from these sources on fictive EEG/MEG

channels reproduced classical auditory evoked waveforms and top-

ographies. In conclusion, the spatiotemporal pattern of activation of

supratemporal auditory areas could be identified on the individual

anatomy using current estimates from intracerebral data. Such detailed

localization approach could also be used prior to epilepsy surgery to

help identify epileptogenic foci and preserve functional cortical areas.

D 2005 Elsevier Inc. All rights reserved.

Keywords: Auditory cortex; Middle-latency components; L1-norm;

Minimum current estimates; EEG; MEG; Stereotactic EEG; Intracranial;

Distributed source method

1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved.

doi:10.1016/j.neuroimage.2005.05.056

* Corresponding author. Present address: LNR-CNRS UMR5816, Bati-

ment B2, Avenue des facultes, F-33405 Talence cedex, France. Fax: +33 5

40 00 25 61.

E-mail address: b.yvert@lnr.u-bordeaux1.fr (B. Yvert).

Available online on ScienceDirect (www.sciencedirect.com).

Introduction

The main human auditory areas are housed in the supratemporal

plane (STP) at the level of the first Heschl’s gyrus (H1) and

surrounding Planum Temporale (PT), superior temporal gyrus

(STG), and occasionally other transverse gyri (H2 and/or H3) when

these are present. Anatomical studies (Brodmann, 1909; Galaburda

and Sanides, 1980; Pandya, 1995; Rivier and Clarke, 1997;

Morosan et al., 2001; Hackett et al., 2001) have indeed subdivided

the STP into several distinct areas, with a general agreement,

supported by intracerebral recordings (Celesia, 1976; Liegeois-

Chauvel et al., 1991), that the primary auditory cortex is located

within the most medial portion of H1, sometimes striding postero-

laterally over Heschl’s sulcus (HS) or even H2 when present

(Rademacher et al., 2001; Hackett et al., 2001). The detailed

spatiotemporal pattern of activation of these multiple areas remains

largely unknown.

In humans, different auditory evoked components are com-

monly recorded using electro- and magnetoencephalography

(EEG, MEG). Although several scalp studies have provided

equivalent dipole sources of auditory evoked components in the

supratemporal plane, very few investigations have been focusing

on precise anatomical locations of auditory areas with respect to

detailed individual gyral and sulcal anatomy (Lutkenhoner and

Steinstrater, 1998; Yvert et al., 2001). In a previous EEG/MEG

study, we reported several STP activities underlying the Pa/Pb

complex of the middle-latency components (between 30 and 60

ms) involving H1/HS and secondary areas in the PT and

supratemporal gyrus (STG) (Yvert et al., 2001). However, scalp

EEG/MEG seldom present sufficient spatial precision and signal-

to-noise ratio to probe earliest (and weakest) activities and

disclose sources very close spatially and greatly overlapping in

time.

The increased spatial resolution of fMRI has allowed to identify

several STP areas involved in auditory perception (Lauter et al.,

1985; Binder et al., 1994; see review by Griffiths and Warren,

2002), although not offering enough temporal resolution to observe

their dynamics on a millisecond time scale.

B. Yvert et al. / NeuroImage 28 (2005) 140–153 141

Intracranial auditory evoked potentials (AEPs) obtained in

epileptic patients having electrodes chronically implanted in the

temporal lobe certainly offer great details of STP activities

(Celesia, 1976; Lee et al., 1984; Liegeois-Chauvel et al., 1991;

Howard et al., 1996, 2000). The anatomic origin of intracerebral

data is generally determined by visual inspections of the curves at

each recording site and by assuming that sources are close to

waveform maxima or polarity reversals occurring between

adjacent contacts. In particular, volume conduction effects are

not taken into account. A drawback of the wealth of details of

intracerebral data is also the difficulty to obtain global and

synthetic views of (generally simultaneous) activities, within and

especially across subjects, where the number and locations of the

electrodes generally differ.

Here, we have adapted to intracerebral data the minimum

current estimates distributed source method (weighted MCE)

originally introduced for scalp data (Matsuura and Okabe, 1995,

1997; Uutela et al., 1999). This method was then used to

localize sources of intracerebral AEPs before 100 ms with

respect to the individual gyral and sulcal STP anatomy in three

patients having several deep multicontact electrodes surrounding

the STP.

Methods

Patients

Signals were recorded using deep multicontact intracerebral

electrodes chronically implanted in the right hemisphere of three

right-handed patients (FY, DC, NG) undergoing presurgical

evaluation of a pharmaco-resistant temporo-mesial epilepsy (aged

23–40, 2 women) and having at least 3 electrode tracks

surrounding H1, PT, and STG. The relatively high number of

electrodes that circumscribed the STP was required to obtain good

localization results. This perquisite was however very rarely

satisfied for most patients recorded in clinical routine at the

hospital had only 1 or 2 deep electrodes exploring the STP, and it

was very seldom that 3 electrodes were implanted in this region. In

the present study, the number of tracks was 7 for subject FY, 7 for

DC, and 6 for NG (Fig. 1). Electrodes were 2-mm-long cylinders

0.8 mm in diameter separated by 1.5 mm of insulation. The three

patients reported no auditory complaints and gave informed

consent to undergo these routine clinical recordings serving as

pre-operative cortical functional mapping and approved by the

ethical committee of the hospital.

Experimental paradigm

Stimuli consisted of 1000 short 1000-Hz pure tones (3-ms rise/

fall times, 50-ms plateau) delivered at 60 dB above hearing

threshold to the ear contralateral to the implanted hemisphere every

200–400 ms (random stimulus onset asynchrony). The three

patients were comfortably seated with the instruction to keep their

eyes open. One patient watched a silenced self-selected movie

during the recordings.

Signals were acquired continuously, bandpass filtered (1–500

Hz), digitized at 2 kHz, and averaged off-line after periods of

artifacts were discarded by careful visual inspection of the raw

data. Contacts showing high level of noise or epileptic spikes were

also discarded so that 59 (FY), 47 (DC), and 32 (NG) channels

were finally considered. In two patients (FY and DC), ground and

reference were taken at intracerebral contacts distant from the

STP. In patient NG, ground and reference were taken on the scalp.

In all cases, signals were average-referenced prior to source

estimation.

Anatomical registration of the electrodes

For each patient, T1-weighted 3D MRIs (voxel size 1 � 1 �1.27 mm) were acquired prior to implantation. Angiography was

used to determine the target electrode positions in Talairach space

as described by Musolino et al. (1990). Because our depth

multicontact electrodes were not visible on MRIs, their final

anatomic positions (which could slightly differ from their target

ones) were determined a posteriori using 2 orthogonal sagittal and

coronal X-ray radiographs obtained at the end of the implantation.

This registration method was described previously (Yvert et al.,

2002): first, the bone contour of the mid-sagittal MRI was matched

to that of the sagittal X-ray view, providing antero-posterior and

inferior–superior coordinates. Then, medio-lateral coordinates

were obtained on the coronal X-ray view. The registration

precision was evaluated to be ¨2 mm in one patient (not included

in this study) for whom MRIs, acquired less than 24 h after

electrode removal, still showed electrode tracks.

The minimum current estimates (MCE) method

The MCE method, as all distributed source methods, makes no

assumption on the number of active regions. A set of candidate

sources with fixed positions and orientations (source domain) is

given a priori, and the strength of each source is estimated so as to

explain the data under some constraints. Let N be the number of

candidate sources, M (<N) the number of electrode contacts, and L

the lead-field matrix, the (i, j) element of which being the potential

created by the jth dipole source with strength 1 onto the ith channel.

Let D be the M-vector of potential values at a given latency, and J

the N-vector of unknown source strengths. The weighted MCE

seeks for the particular solution J* verifying

jjWJ4jj1 ¼ MinJ

jjWJ jj1=LJ ¼ D��; ð1Þ

where ||&||1 denotes the L1-norm, and W is an N*N diagonal matrix

of weighting factors precluding possible bias favoring sources

close to the sensors:

Wj ¼ffiffiffiffiffiffiffiffiffiffi~i

L2ij

r

By writing the strength of the jth source

Jj ¼ Jþj � J�j ;

where J j+ and J j

� are positive values, and using the singular value

decomposition of L = USVt, Eq. (1) can be written as the following

linear programming (LP) problem with 2*N unknowns Jj+ and Jj

�:

Minimize :~j

Wj Jþj þ J�j

� �;

subject to the M constraints:

SVt Jþ � J�ð Þ ¼ UtD: ð2ÞIt should be noted that the number of constraints actually

determines the maximum number of non-zero solution values in

Fig. 1. Visualization of the source domain and position of electrode contacts in each subject. (A) Subject FY. (A1) Sagittal view of the right-hemisphere

midgray surface showing the position of the electrode tracks. (A2) The extracted STP region is surrounded by several multicontact electrodes, 3 of which

contained most of the signal: electrodes H and T located just below the posterior and anterior part of the STP, respectively, and electrode N above the STP. (A3)

Distribution of candidate dipole sources with respect to electrode positions. (B and C) Source domain and electrode contacts for subject DC and NG,

respectively. For subject NG, two contacts of an orbito-frontal electrode track (FO_) were included even though these contacts showed no AEPs. Electrodes

labeled in color corresponded to those showing most signal and were used to build the spatiotemporal maps in Fig. 3. The orientation of the source domain is

the same for the 3 subjects: m, medial; s, superior; a, anterior. Scale: electrode contacts are 2-mm-long cylinders separated by a gap of 1.5 mm.

B. Yvert et al. / NeuroImage 28 (2005) 140–153142

vector J*. In practice, a regularization variant of this formulation is

required to obtain solutions more stable with respect to noise

(Matsuura and Okabe, 1997; Uutela et al., 1999; Fuchs et al.,

1999). Here, we used the strategy proposed by Uutela et al. (1999),

which consists in reducing the number of constraints in Eq. (2).

The lowest rows of the linear system (Eq. (2)) indeed correspond to

the smallest singular values and require high coefficients of J to

meet the non-zero right-hand side. Hence, removing these rows

stabilize the solutions. In practice, we chose a number of

constraints between M/4 and M/2: 26 for FY, 15 for DC, and 16

for NG. The number of constraints was chosen so as to obtain a

tradeoff between the number of degrees of freedom of the solution

(number high enough to allow several possible simultaneous

sources) and the stability of the solution (number small enough to

obtain a solution stable with respect to noise). The LP problems

were solved using the CPLEX package (ILOG, Homburg,

Germany).

Construction of the source domain

For each patient, the white matter and pial surfaces of the

implanted hemisphere were segmented using the Freesurfer tool set

freely available at http://surfer.nmr.mgh.harvard.edu/. The inter-

mediate Fmidgray_ surface located midway between white and pial

surfaces was then computed (Fig. 1A1). The midgray STP region

including Heschl’s gyri, PT, and STG was extracted, and the

corresponding mesh was simplified to contain approximately 80

triangles/cm2 (Figs. 1A2, B1, C1). This portion of cortical surface

constituted the source domain: one source was positioned on each

mesh node with a fixed orientation perpendicular to the local

surface (Figs. 1A3, B2, C2). The number of nodes of the source

domain was 1128 for FY, 1013 for DC, and 1535 for NG. The

average distance between neighboring sources was 1.7 mm. The

source domain was chosen on the basis of anatomical data

available from the literature (Brodmann, 1909; Galaburda and

B. Yvert et al. / NeuroImage 28 (2005) 140–153 143

Sanides, 1980; Rademacher et al., 1993, 2001; Pandya, 1995;

Rivier and Clarke, 1997; Morosan et al., 2001; Hackett et al., 2001)

so as to include the regions known to house the main human

auditory areas within the superior temporal plane and supra-

temporal gyrus. This domain was thus limited medially by the

bottom of the insula, laterally by the upper bank of the superior

temporal sulcus, caudally by the superior end of the Planum

Temporale, and rostrally by the anterior limit of Heschl’s gyrus. In

patient NG, a more anterior part of the supratemporal gyrus was

considered because several more anterior electrodes were present

among which electrode FT_ showed significant signal on its

contacts. By contrast, subject FY had only one anterior electrode

track (FJ_) which showed no signal, and subject DC had no anterior

electrodes. Figs. 1A–C show the source domain with respect to the

electrode contacts for all 3 subjects. Each electrode track was

labeled by a letter (e.g., FH_ for the electrode track close to Heschl’sgyrus), and each contact on the track was numbered starting from

F1_ for the most medial contact. As seen in Fig. 1, several tracks

surrounded the supratemporal plane for each subject.

Lead-field computation

The intracerebral lead-fields were calculated with a single-layer

homogeneous spherical model (conductivity r = 0.45 S/m) best

fitting the inner skull surface lateral to the STP. For a given source,

(r,h,/) denotes spherical coordinates with origin at the center of

the model and z axis passing through the source, and Q and r0stand for the dipole moment and position, respectively. The

potential V(r) created at any intracerebral location r = (r,h,/)

within the model is given by (Bertrand et al., 1991):

V ¼ U þ Vs; ð3Þ

where

Vs ¼1

4prr� r0ð Þ IQjr� r0j3

is the solution that would be created by the source in an infinite

medium of conductivity r, and

U r; h;/ð Þ ¼ 1

4pr

XVn ¼ 1

nþ 1ð Þ

� rn � 10

R2n þ 1rn QrPn coshð Þ þ QtP

1n coshð Þcos/

�ð4Þ

In Eq. (4), R is the radius of the spherical model, Qr and Qt are

the radial and tangential components of the source moment, and Pn

and P1n denote Legendre and associated Legendre polynomials,

respectively.

Solution stabilization using bootstrap reaveraging of the single

trials

In order to check whether a solution was stable with respect to

background noise, we repeated the source estimation on 30 different

average responses obtained by bootstrap reaveraging of the initial

single trials. This method was described previously (Yvert et al.,

2002): let NS be the number of available single trials. Thirty sets of

NS single trials were randomly drawn with replacement from the

original set of NS single trials (NS = 1000). Averaging trials within

each set lead to 30 bootstrap average responses spanning the

intrinsic variability of the average data, without assumption on the

noise structure. At each time sample between 0 and 100 ms, STP

sources were then estimated using the weighted MCE method for

each bootstrap average, leading to 30 bootstrap solutions. These 30

solutions were then averaged to obtain a first solution J1. Next, only

nodes displaying stable solution values across bootstraps were

considered as reliably active, although allowing that the activity

could locally jump a few nodes across bootstraps: to do so, a node

value of J1 was considered unstable and thus zeroed if there was at

least one bootstrap for which no node within a geodesic distance of

7.5 mm was active (i.e. non-zero). Then, a threshold was applied so

that nodes with values below 15% of the maximum value of the

map were zeroed. This lead to the modified solution J1m. Finally,

J1m was globally rescaled by a scalar k in order to minimize the

quadratic error between original data D (mean of all bootstrap

averages) and reconstructed data: Jfinal = k J1m with k = (D, LJ1m) /

(LJ1m, LJ1m) and (,) being the dot product.

Evaluation of the inverse procedure

Simulations were used to test the inverse procedure on 100

pairs of two simultaneous sources located randomly in the source

domain and separated by at least 15 mm. For each source pair, the

corresponding intracerebral data created by the 2 sources (with

amplitude 1 nA) were computed at the intracranial electrodes using

Eq. (3), and the inverse procedure described above was used to

retrieve the sources. Four situations were tested with increasing

perturbations on the data: (1) data were left unchanged, (2) thirty

different random Gaussian noises (signal-to-noise ratio SNR = 25,

which corresponded to typical values encountered in our record-

ings) were added to these data to create 30 different data sets,

mimicking 30 bootstrap averages, (3) the two sources were shifted

by 1.7 mm (i.e. 1 internode distance) and tilted by 5- prior to

computing the data, and random Gaussian noise was added as

before, (4) the two sources were shifted by 3.5 mm and tilted by

10- prior to computing the data, and random Gaussian noise was

added. Source shifts mimicked possible registration error of the

electrodes, and source tilts mimicked possible errors in the

estimation of the source orientations on the cortical surface. For

each solution, patches of activity were defined as clusters of active

neighboring nodes. Smoothed patches were also defined as clusters

of active neighboring nodes after the solution was spatially

smoothed with a Gaussian filter of radius 7.5 mm (r = 2.5 mm,

coefficients zeroed beyond 3*r).

Results

Evaluation of the localization method

Results of the simulations are shown in Fig. 2, which can be

read from left to right to see the effect of increasing perturbations.

Fig. 2A shows three examples of source reconstructions for 3

different pairs of dipoles. When data were not perturbed and the

source domain contained the sources used to generate the data, the

inverse procedure reconstructed perfectly these two sources in

96% of the cases. It should be noted that 100% perfect

reconstruction was achieved in absence of regularization when

no constraints were removed in Eq. (2), but then these solutions

became very unstable with respect to noise so that this approach

could not be used in practice. As data were increasingly perturbed,

Fig. 2. Evaluation of the inverse procedure. One hundred pairs of dipoles distant by at least 15 mm were chosen randomly on the source domain. As explained

in the Methods section, 4 types of perturbation were applied to the data before the sources were reconstructed using the MCE method. From left to right, the

amount of perturbation increases. (A) Examples of source reconstruction for 3 dipole pairs and the 4 perturbations (original source positions are marked by red

dots). Orientation as in Fig. 1: m, medial; s, superior; a, anterior. (B–E) Histograms showing the number of active patches reconstructed (B), the distance

between each reconstructed patch and the nearest actual source position (C), the number of active patch after spatial smoothing (D), and the number of

smoothed patches close to each dipole source (E). Overall, these results show that, as perturbations increased, the method did not tend to generate multiple

spurious phantom sources. Rather, sources were either correctly reconstructed (usually by several neighboring patches) or not reconstructed at all. It should be

noted that the position of the sources with respect to the electrodes was random and sources could be reconstructed even at remote sites with respect to the

electrodes (e.g., source in the Planum Temporale in A2 or A3).

B. Yvert et al. / NeuroImage 28 (2005) 140–153144

B. Yvert et al. / NeuroImage 28 (2005) 140–153 145

the number of patches of activity increased (Figs. 2A and B).

However, these patches remained close to each other, meaning that

each source became reconstructed by several small neighboring

patches (e.g., see Figs. 2A2 and A3). This is further assessed in

two ways. First, the distances between each patch and the nearest

dipole remained acceptable even for the strongest perturbations

(Fig. 2C). Second, after spatially smoothing the solutions, most

reconstructions (88%) showed actually only 1 or 2 smoothed

patches of activity for the strongest perturbation (Fig. 2D), and

either 1 or zero patch was reconstructed next to each dipole (Fig.

2E). This important result shows that the method was robust in the

sense that, as data became more and more perturbed, the method

did not tend to generate multiple spurious phantom sources.

Rather, sources were either correctly reconstructed (usually by

several neighboring patches) or not reconstructed at all. It should

be noted that the position of the sources with respect to the

electrodes was random and sources could be reconstructed even at

remote sites with respect to the electrodes (e.g., source in the

Planum Temporale in Figs. 2A2 or A3).

Spatiotemporal mapping of intracranial AEPs

Intracranial AEPs were characterized by a complex succession

of multiple components at several electrode contacts. This is

illustrated in Fig. 3 for all subjects, where the 30 bootstrap

averages are superimposed to show the intrinsic variability of the

data. As shown in Figs. 3B, D, and F, more synthetic visualization

of the data recorded simultaneously on several channels was

achieved by displaying spatiotemporal maps (Badier and Chauvel,

1995). Only channels showing most activity were included in these

maps. These representations showed a succession of activities,

either as peaks occurring simultaneously on several contacts or as

polarity reversals (PRs) between adjacent contacts of the same

track. Based on these maps, we divided the activity into four

distinct periods for each subject (gray bars below each map): (1) a

weak initial activity occurring at 16–19 ms on medial H contacts

and characterized by a PR in subjects FY and DC; (2) then, a PR at

20–25 ms involving slightly different medial H contacts and

spreading monopolarly on other tracks (e.g., tracks N and T in

subject FY and NG); (3) the 30–50-ms time range then showed a

more complex pattern of evoked activity, which remains relatively

stable in this range and showed several simultaneous PRs in

subjects FY (between H5 and H6 and between H6 and H7) and DC

(between H2 and H3 and between H3 and H4), suggesting several

simultaneous sources; (4) between 50–60 ms and 80–100 ms,

these complex activities were still present although with a reversed

polarity on most recording contacts. It is interesting to note that

these 4 time periods of activity actually correspond fairly well to

the time ranges of known scalp auditory evoked components: (1)

P0, (2) Na, (3) Pa/Pb, and (4) N100. In an attempt to link our

results with classical scalp data, we used a parallel terminology to

that of scalp components with the ?iX subscript standing for

?intracerebralX: P0i, Nai, Pai/Pbi, and N100i.

Estimation of underlying sources

The sources of the intracerebral AEPs were estimated at all

latencies between 0 and 100 ms in each subject, using the weighted

MCE method described above. Results are presented in Figs. 4–6

for each time periods separately. For sake of clarity, the sources are

shown at key latencies corresponding to either the maxima of the

evoked components (Fig. 4 for P0i and Nai) or at several latencies

in order to describe the detailed evolution of the activity in the

supratemporal plane in a movie-like fashion (Fig. 5 for Pai/Pbi and

Fig. 6 for N100i). These latencies are indicated by vertical dashed

lines on the corresponding spatiotemporal maps in Fig. 3. Then, in

order to allow the correspondence between localization maps and

electrode data, AEPs were color-coded at each contact next to the

localization results. Although individual STP anatomy and source

localization results were variable across subjects, we found a

common spatiotemporal pattern of activity.

As seen in Fig. 4, the earliest activity (P0i) occurred between 16

and 19 ms, with a single source localized in the postero-medial

portion of HS in 2 subjects (FY, DC) and of H1 in subject NG and

pointing outward of the midgray surface. Then, the Nai (20–25

ms) stemmed from an inward source in H1/HS at a different

location than P0i in all three subjects.

Next, after 25 ms, we observed the progressive activation of

several sources in the supratemporal plane, having outward

orientations. As seen in Fig. 5, activity propagated medio-laterally

and postero-anteriorly as latency increased during the Pai/Pbi time

range. For subject FY, sources tended to overlap in time. At 28.5

ms, activity could be found in medial HS, PT, and also already in

the STG. Then, the STG tended to become silent while activity left

HS and involved only the PT around 42.5 ms, before propagating

to a more anterior part of HS and to the STG again. It can be noted

that this scheme of activity was very consistent with the activities

recorded on the electrodes (see maps in Fig. 3B and color-coded

data next to the source reconstructions). At 49 ms, these two last

regions remained active, while the PT has become silent. An

inwardly oriented source then started to be active in the posterior

part of H1, indicating the beginning of the N100i. For subject NG,

an early activation of the STG was also found at 28.5 ms. Then, at

32 ms, activity was located in the postero-medial part of H2 and

then propagated laterally along H2. At about 40 ms, the anterior

part of H2 was solely active as was the PT for subject FY. Finally,

the STG became active around 45 ms at the level of the anterior

end of HS. The same type of spatiotemporal scheme was also

observed in subject DC between 30 and 41 ms, with activity first

involving HS and H2 at 30 ms and then also the anterior part of HS

and H3 at 35 ms. At 40 ms, activity was predominant in lateral

areas (H2 and H3). It can be noted that this lateral progression of

activity could also be seen on the spatiotemporal map presented in

Fig. 3F. By 46–50 ms, the lateral activity at the anterior part of H3

remained, while activity in medial H1 started again characterizing

the beginning of the N100i. For this subject, some polarity

reversals, such as that at medial H contacts at the beginning of

the N100i, were reconstructed as a few neighboring inward and

outward sources and not as single patches. This was due to the fact

that electrode H actually crossed the supratemporal plane in several

places (see Fig. 1B1) generating very strong lead-fields in these

places (note also the high signal-to-noise ratio of the data for this

subject). Despite the presence of these singularities, the weighted

MCE method actually managed to reconstruct the sources fairly

well.

Finally, as seen in Fig. 6, the supratemporal sources of the

N100i showed comparable activation patterns in subjects FY and

DC (Fig. 6, left and right columns). This activity was characterized

by sources oriented inwardly, beginning around 52–55 ms in the

postero-medial part of H1/HS. Then, activity propagated to the PT

in subject FY and in H2 for subject DC around 58–60 ms before

reaching more lateral regions (STG for FY and H3 for DC) around

g. 3. Bootstrap averages (top row) and spatiotemporal maps (bottom row). (A, C, E) Superimposed bootstrap averages for 5 recording contacts of H, N, a electrodes in subject FY, NG, and DC, respectively.

bject DC showed a very high signal-to-noise ratio. (B, D, F) Spatiotemporal maps showing the evoked AEPs on all contacts of the 3 electrodes showing st signals for subjects FY, NG, and DC, respectively.

he 4 time ranges considered are represented by horizontal gray bars below each map. Vertical lines indicate the latencies for which the localization res are shown in Figs. 4–6.

B.Yvert

etal./NeuroIm

age28(2005)140–153

146

Fi

Su

T

nd T

mo

ults

Fig. 4. Sources of the earliest components P0i and Nai. Source reconstruction are color mapped for all three subjects at the latency corresponding to the peak of the

P0i and Nai components as seen on the spatiotemporal map (see vertical lines in Fig. 3). Below each source map, the AEPs recorded at the corresponding latency

are represented on each electrode contact as a color scale (dark colors correspond to small potential values). The position of each source of the domain is

represented as a dot. Scale: electrode contacts are 2-mm-long cylinders separated by a gap of 1.5 mm. Orientation as in Fig. 1: m, medial; s, superior; a, anterior.

B. Yvert et al. / NeuroImage 28 (2005) 140–153 147

70–100 ms. For these two subjects, the areas involved in the

N100i were very similar to those involved in the Pai/Pbi complex.

This postero-anterior and medio-lateral propagation of activity was

also clearly seen in subject NG between 65 and 95 ms, where it

successively stemmed from medial HS, anterior HS, and the STG

at the level of the tip of H1 (Fig. 6, middle column). A more

posterior region of the STG was also active between 65–75 ms. In

this subject, areas involved in the N100i were generally more

anterior than those involved in the Pai/Pbi complex.

Reconstruction of fictitious scalp AEPs and AEFs

Comparing intracerebral and scalp AEPs has shown to be very

difficult for two main reasons. First, for practical reasons, it is rare

that both types of signals can be recorded simultaneously (Godey

et al., 2001). Second, intracerebral AEPs are characterized by

multiple local features highly dependent on the location of the

recording contacts and are thus difficult to relate to the relatively

simple succession of well-known scalp components. Although

scalp data were not available for our patients, we checked that the

sources reconstructed from intracerebral data were compatible with

classical scalp components. For that purpose, we reconstructed

fictitious scalp components produced by the sources estimated

from intracerebral data. For each subject, fictitious scalp AEPs

were computed at 32 electrodes extending the 10–20 system using

a classical 3-shell spherical model. For subject FY, we also

reconstructed fictive AEFs on the BTi 37-channel gradiometer

configuration covering the right hemisphere. When reconstructing

AEPs, the source configurations obtained in the right hemisphere

using the weighted MCE were symmetrized in the other hemi-

sphere to account for the activity of both hemispheres. Because

MEG is more focal than EEG, this was not done for AEFs. As

shown in Fig. 7, these reconstructed data resembled classical scalp

evoked responses (Na, Pa, Pb, N100).

Discussion

Up to now, source analysis methods were exclusively used to

analyze scalp data. Although previous simulations have shown that

source position and strength can be estimated from intracerebral

data by computing the mean electric field and center of energy

(Lemieux and Leduc, 1992), such method can only be considered

in the case of a single active source, which is rarely the case. The

first purpose of our study was to illustrate that scalp methods can

also be used to estimate sources of intracerebral data, despite the

fact that dipole approximation might not be optimal in this case

where sources are close to the recording contacts (Church et al.,

Fig. 5. Sources of the components Pai/Pbi. Source reconstructions are color mapped for all three subjects at several latencies spanning the Pai/Pbi time range

(see vertical lines in Fig. 3). On the right of each source map, the AEPs recorded at the corresponding latency are represented on each electrode contact as a

color scale. The position of each source of the domain is represented as a dot. Scale: electrode contacts are 2-mm-long cylinders separated by a gap of 1.5 mm.

Orientation as in Fig. 1: m, medial; s, superior; a, anterior.

B. Yvert et al. / NeuroImage 28 (2005) 140–153148

1985). The second goal of this paper was to use this method in

order to localize the supratemporal sources of intracerebral AEPs.

We found that the weighted MCE method could be used to

reconstruct sources from intracerebral data on the individual gyral

and sulcal anatomy and that this method was quite robust when

data were increasingly perturbed: sources tended to be either well

reconstructed or not reconstructed at all. We also tested the effect

of removing several electrodes on the accuracy of the solution. We

found that the solutions were still reliable after electrode removal.

Although the precision of the localization decreased as perturbation

increased, solutions remained robust even when electrodes show-

ing the maximum signals were removed.

We compared the MCE method with other distributed source

methods such as the weighted L2-norm (Hamalainen and

Ilmoniemi, 1994) and the LORETA (Pascual-Marqui et al., 1994)

methods combined with Tikhonov regularization (and determina-

tion of optimal lambda with the L-curve, see Hansen and O’Leary,

1993). In particular, we tested how these different methods

performed when 2 patches of increasing size were reconstructed

(Fig. 8). In theory, the MCE method is expected to perform less

reliably in a situation with large distributed sources since this

method seeks to reconstruct focal sources (Uutela et al., 1999).

Two patches of dipoles were considered having amplitudes

following a Gaussian distribution in space (in the geodesic sense)

of radius equal to 3*r (r = 0, 1, 2, 3, and 4 mm). No perturbations

were introduced. As the size of the patch increased, we found that

the L1 method tended to reconstruct several nearby focal sources,

the strongest of which being located near the center of the patches.

Fig. 6. Sources of the component N100i. Source reconstructions are color mapped for all three subjects at several latencies spanning the N100i time range (see

vertical lines in Fig. 3). On the right of each source map, the AEPs recorded at the corresponding latency are represented on each electrode contact as a color

scale. The position of each source of the domain is represented as a dot. Scale: electrode contacts are 2-mm-long cylinders separated by a gap of 1.5 mm.

Orientation as in Fig. 1: m, medial; s, superior; a, anterior.

B. Yvert et al. / NeuroImage 28 (2005) 140–153 149

We found that both L2-based methods lead to less reliable

solutions, with ample fantom sources distant from the actual patch

positions.

We also tested whether the localization results obtained at the

different latencies of the auditory evoked responses (Figs. 4–6)

could be summarized using a set of fixed dipole sources located in

the center of the main patches of activity, the amplitude of which

being estimated by a least-square fit. We found that the accuracy of

amplitude estimation strongly depended on the level of colinearity

between the lead-fields of each fixed source. Accurate reconstruc-

tion could be achieved only when all lead-fields were nearly

orthogonal two-by-two. However, when a pair of lead-fields had an

angle smaller than 60-, then the amplitude of the corresponding

sources tended to become strongly correlated with opposite polar-

ities. This problem, which has been raised previously (Lutkenhoner,

2003), could not be alleviated using a regularized version of the

least-square fit by removing the smaller singular values.

For all these reasons, we used the weighted MCE method and

found that individual spatiotemporal functional mapping of supra-

temporal auditory areas could be achieved in subjects having

several multicontact electrodes surrounding the STP.

Our localization results may be paralleled to the anatomical

organization of the supratemporal auditory areas. Detailed parcel-

lation of the supratemporal plane has been achieved in monkeys,

where up to 13 areas organized in 3 major regions have been

identified: the core line region, containing the primary auditory

cortex (PAC) and two more rostral regions, is surrounded by a belt

of 8 distinct areas, which in turn is boarded laterally by two areas

forming the parabelt region in the STG (Pandya, 1995; Rau-

schecker, 1998; Kaas and Hackett, 2000; Kaas et al., 1999). A

similar organization has been described in humans. Recent data

indicate that the human PAC is also contained within a core region

composed of 3 areas along H1, the denser one being located in the

first postero-medial third of H1 (Galaburda and Sanides, 1980;

Rivier and Clarke, 1997; Morosan et al., 2001), often straddling

over H2 when several Heschl’s gyri are present (Rademacher et al.,

1993, 2001; Hackett et al., 2001). Like in monkeys, this core

region is surrounded by a belt of secondary areas medially in the

circular sulcus and laterally over additional Heschl’s gyri (when

present), the PT, and the STG (Galaburda and Sanides, 1980;

Rivier and Clarke, 1997). Functional imaging studies have also

reported several sites of activation in the supratemporal plane

(depending on the stimulus types), mostly involving Heschl’s gyri,

the PT, and the STG, which are the regions we found active here

(Lauter et al., 1985; Binder et al., 1994; Scheich et al., 1998; Belin

et al., 1999; Lockwood et al., 1999; Hashimoto et al., 2000;

Fig. 7. Fictive scalp data reconstructed from intracerebral sources. Topographies of reconstructed scalp Na, Pa, and N100 components are shown on the left

panel: average EEG across all subjects (top row), as well as individual EEG (middle row) and MEG (bottom row) of subject FY. Panels on the right show

reconstructed evoked responses on both electrodes and both MEG sensors circled on the maps (red: averaged EEG, black: individual EEG and MEG). Classical

evoked responses Na, Pa, Pb, and N100 can be recognized. For EEG, average reference was used.

B. Yvert et al. / NeuroImage 28 (2005) 140–153150

Talavage et al., 2000, 2004; Wessinger et al., 2001; Griffiths and

Warren, 2002; Hall et al., 2002; Schonwiesner et al., 2002). These

studies often report activation of the full length of H1 from its

postero-medial end to its antero-lateral tip. Here, we mostly found

activity in the postero-medial part of H1. Activity in the antero-

lateral was observed only in subject NG. More anterior source

locations have been found for more cognitive components tasks

such as the MMN (Csepe et al., 1992). Furthermore, fMRI studies

also report activities on the medial side of H1 in the circular sulcus.

Here, we did not find sources in this region near the insula possibly

because this secondary area of the belt is not activated by simple

tone stimuli. It should be noted that most dipole localization studies

using MEG data also report source activity in H1 and the PT but

not in the circular sulcus (e.g., Pantev et al., 1995; Lutkenhoner

and Steinstrater, 1998; Gutschalk et al., 1999; Herdman et al.,

2003; Yvert et al., 2001). These differences in the localization of

active areas between fMRI studies and our study could stem from

the following facts: (1) we used a very simple tone-evoked

protocol, while most fMRI studies use different paradigms and

stimuli; (2) while fMRI measures variations of blood parameters in

the microvascularization, AEPs directly reflect the neuronal

activity; (3) fMRI integrates activation over several seconds, while

the present localization results reflect activity on a millisecond time

scale; and (4) fMRI signal is not necessarily related to transient

evoked responses only but also to more sustained oscillatory

activities not strictly phase-locked to the stimuli (Logothetis et al.,

2001). Moreover, human fMRI studies and animal multi-unit

recordings using pure tones do not show as focal activation as

those found in the present study, but rather activation over more

extended patches (even on the single subject level without spatial

smoothing). The more focal activity patches found in our study

should be attributed to the MCE method which seeks to reconstruct

focal sources. Indeed, as illustrated in Fig. 8, using the MCE

method, larger patches are reconstructed as several nearby focal

sources, the strongest of which being located near the center of the

patches.

Here, despite an important inter-subject variability, a common

scheme of spatiotemporal activity could be highlighted. Four time

ranges were considered, corresponding to the classical P0, Na, Pa/

Pb, and N100 scalp components.

First, the earliest cortical activity P0i occurring around 16–19

ms, previously described by several authors (Celesia, 1976, Lee et

al., 1984; Liegeois-Chauvel et al., 1991; Howard et al., 1996),

could be localized in the medial portion of HS for two subjects and

H1 in subject NG. This area likely corresponds to the PAC for it is

the first one active. Although an earlier cortical activity has been

previously described around 10 ms using click stimuli (Celesia,

1976; Steinschneider et al., 1999), we did not observe this response

in any of our patient, likely because we used tone burst stimuli

which elicit weaker responses than clicks. Second, another area

also located in the medial H1/HS region was responsible for the

Nai component until about 25 ms. It is possible that this area also

belongs to the PAC, although it was found to be different than the

area involved in the P0i. Indeed, like in monkeys, three areas have

been described to form the PAC in humans (Morosan et al., 2001).

Third, during the Pai/Pbi complex, up to 4 areas became active

with a postero-anterior and medio-lateral propagation of activity in

H1/HS, PT or H2, and then more anterior part of HS and STG or

H3. Given the anatomical description of auditory cortical areas in

monkeys and humans summarized above and given that the belt

region receives projections from the PAC (Brugge et al., 2003), it is

possible that the Pai/Pbi components reflect the activation of

several areas of the lateral belt region and possibly also of the

parabelt in the STG. These regions correspond to areas PaAi,

Fig. 8. Comparison between the MCE method and L2-based methods (weighted MNE and LORETA regularized using the Tikhonov approach and the L-curve)

for activity patches of increasing sizes. Two patches of dipoles were considered, each having amplitudes following a Gaussian distribution in space (in the

geodesic sense) of radius equal to 3*r (r = 0, 1, 2, 3, and 4 mm). The centers of the patches are indicated by red crosses. No perturbations are introduced. The

MCE method tends to reconstruct several neighboring focal sources, the strongest of which being close to the centers of the patches. The L2-based methods

tend to generate fantom sources a remote distance from the patches.

B. Yvert et al. / NeuroImage 28 (2005) 140–153 151

PaAe, and Tpt described by Galaburda and Sanides (1980, Fig. 1)

and to areas PA, LA/STA, and AA identified by Rivier and Clarke

(1997, Fig. 10). Activities in the PT for subject FY corresponded to

activities in H2 in the two other subjects, suggesting a similar

functional role of these structures. Finally, during the N100i time

range, the same areas were active in two subjects, whereas in

subject NG, the sources were found slightly more anteriorly.

The present results confirm and extend in more details previous

EEG/MEG findings reporting 3 different active areas underlying the

Pa/Pb complex (Yvert et al., 2001). However, in this previous scalp

study, the earliest components (P0 and Na) could not be detected

and the corresponding source locations not estimated. Actually,

achieving precise anatomical localization of the sources of the

earliest auditory evoked cortical components occurring before 30

ms has turned to be very difficult using scalp data. A recent MEG

study, using a very high number of click stimuli (8500) to obtain a

sufficient signal-to-noise ratio, suggested that the postero-medial

part of H1 was solely responsible for the earliest ascending phase of

the Pa component already at 20 ms (Lutkenhoner et al., 2003). Here,

using pure tones, we found that the Pa component started after 25

ms. This latency difference may likely be due to the fact that clicks

are more broadband stimuli recruiting a larger amount of cells and

leading to smaller latencies than single-frequency pure tone stimuli

(e.g., see Fig. 1 of Woods et al. (1995) showing latency differences

of several ms). To our knowledge, however, detailed source

localization of the P0 and Na components elicited by pure tones

has never been presented. Mapping analysis of click-evoked AEPs

has suggested that Na and Pa likely stemmed from different

generators (Deiber et al., 1988) and that Na sources were localized

deeper in the brain than Pa sources, possibly even from diencephalic

structures. Here, we propose that the Na component reflects the

activation of a single source from the postero-medial part of H1 and

that, by contrast, the Pa already stems from the simultaneous

activation of several sources including lateral areas such as the PT

and the STG. Thus, a single dipole model is a good approximation

of the auditory activity in the STP only at the earliest stage of the

evoked response (P0/Na). In particular, after 25 ms, several sources

become simultaneously active and would not be perfectly disclosed

with this simplified model.

A pioneer localization study by Scherg and Von Cramon (1986)

showed that middle-latency (19–40 ms) and long-latency (45–200

ms) AEPs recorded on a near-coronal arrangement of electrodes

B. Yvert et al. / NeuroImage 28 (2005) 140–153152

could be well explained by the activity, in each hemisphere, of a

radial and a tangential stationary sources that were maximally

active at different latencies: N19, P30, and N100 for the tangential

source and N27, P39, and P100 for the radial source. It was

suggested that the tangential source could mimic the activity of

Heschl’s gyrus and that the radial source could reflect the activity

of the STG. This hypothesis was further supported by an MEG

analysis of steady-state and middle-latency components reporting

two underlying sources: a tangential source in postero-medial H1

and another source in the PT (Gutschalk et al., 1999). Our

localization results extend these studies by showing that the

number of active sources was actually greater than 2 thereafter 25

ms and that 3 to 4 areas (including H1, PT, and STG) then

overlapped their activities during both the Pa/Pb and the N100 time

ranges. These findings support that source localization from

intracerebral EEG data using distributed source methods, which

do not make any assumption on the number of sources, can help

distinguish between several simultaneously active regions which

may be difficult to separate using MEG/EEG (see also Lutkenh-

oner (2003) on the limit of MEG to separate close sources).

In conclusion, the spatiotemporal pattern of activation of

supratemporal auditory areas could be identified on the individual

anatomy using current estimates from intracerebral auditory

evoked potentials. The localization approach described in this

study also offers a new way to map functional areas prior to

cortical resections considered to cure pharmaco-resistant epilep-

sies. Furthermore, such source localization approach could also be

useful for precise sulcal and gyral identification of epileptic foci

prior to surgery.

References

Badier, J.M., Chauvel, P., 1995. Spatio-temporal characteristics of

paroxysmal interictal events in human temporal lobe epilepsy.

J. Physiol. (Paris) 89, 255–264.

Belin, P., Zatorre, R.J., Hoge, R., Evans, A.C., Pike, B., 1999. Event-related

fMRI of the auditory cortex. NeuroImage 10, 417–429.

Bertrand, O., Thevenet, M., Perrin, F., 1991. 3D finite element method in

brain electrical activity studies. In: Nenonen, J., Rajala, H.M., Katila,

T. (Eds.), Biomagnetic localization and 3D modelling, Technical

Report TKK-F-A689, Helsinki University of Technology, Helsinki,

pp. 154–171.

Binder, J.R., Rao, S.M., Hammeke, T.A., Yetkin, F.Z., Jesmanowicz, A.,

Bandettini, P.A., Wong, E.C., Estkowski, L.D., Goldstein, M.D.,

Haughton, V.M., Hyde, J.S., 1994. Functional magnetic resonance

imaging of the human auditory cortex. Ann. Neurol. 35, 662–672.

Brodmann, K., 1909. Vergleichende Lokalisationslehre der Grosshirnrinde

in Ihren Prinzipien Dargestellt auf Grund des Zellaufbaus. Barth,

Leipzig.

Brugge, J.F., Volkov, I.O., Garell, P.C., Reale, R.A., Howard III,

M.A., 2003. Functional connections between auditory cortex on

Heschl’s gyrus and on the lateral superior temporal gyrus in

humans. J. Neurophysiol. 90, 3750–3763.

Celesia, G.G., 1976. Organization of auditory cortical areas in man. Brain

99, 403–414.

Church, P., Leduc, A., Beique, R.A., 1985. Sensitivity analysis of depth

EEG electrodes to dipolar electric sources. IEEE Trans. Biomed. Eng.

32, 554–560.

Csepe, V., Pantev, C., Hoke, M., Hampson, S., Ross, B., 1992. Evoked

magnetic responses of the human auditory cortex to minor pitch

changes: localization of the mismatch field. Electroencephalogr. Clin.

Neurophysiol. 84, 538–548.

Deiber, M.P., Ibanez, V., Fischer, C., Perrin, F., Mauguiere, F., 1988.

Sequential mapping favours the hypothesis of distinct generators for Na

and Pa middle latency auditory evoked potentials. Electroencephalogr.

Clin. Neurophysiol. 71, 187–197.

Fuchs, M., Wagner, M., Kohler, T., Wischmann, H.A., 1999. Linear and

nonlinear current density reconstructions. J. Clin. Neurophysiol. 16,

267–295.

Galaburda, A., Sanides, F., 1980. Cytoarchitectonic organization of the

human auditory cortex. J. Comp. Neurol. 190, 597–610.

Godey, B., Schwartz, D., de Graaf, J.B., Chauvel, P., Liegeois-Chauvel, C.,

2001. Neuromagnetic source localization of auditory evoked fields and

intracerebral evoked potentials: a comparison of data in the same

patients. Clin. Neurophysiol. 112, 1850–1859.

Griffiths, T.D., Warren, J.D., 2002. The planum temporale as a computa-

tional hub. Trends Neurosci. 25, 348–353.

Gutschalk, A., Mase, R., Roth, R., Ille, N., Rupp, A., Hahnel, S., Picton, T.,

Scherg, M., 1999. Deconvolution of 40 Hz steady-state fields reveals

two overlapping source activities of the human auditory cortex. Clin.

Neurophysiol. 110, 856–868.

Hackett, T.A., Preuss, T.M., Kaas, J.H., 2001. Architectonic identification

of the core region in auditory cortex of macaques, chimpanzees, and

humans. J. Comp. Neurol. 441, 197–222.

Hall, D.A., Johnsrude, I.S., Haggard, M.P., Palmer, A.R., Akeroyd, M.A.,

Summerfield, A.Q., 2002. Spectral and temporal processing in human

auditory cortex. Cereb. Cortex 12, 140–149.

Hamalainen, M.S., Ilmoniemi, R.J., 1994. Interpreting magnetic fields of

the brain: minimum norm estimates. Med. Biol. Eng. Comput. 32,

35–42.

Hansen, P.R., O’Leary, D.P., 1993. The use of the L-curve in the

regularization of discrete ill-posed problems. SIAM J. Sci. Comput.

14, 1487–1503.

Hashimoto, R., Homae, F., Nakajima, K., Miyashita, Y., Sakai, K.L., 2000.

Functional differentiation in the human auditory and language areas

revealed by a dichotic listening task. NeuroImage 12, 147–158.

Herdman, A.T., Wollbrink, A., Chau, W., Ishii, R., Ross, B., Pantev, C.,

2003. Determination of activation areas in the human auditory cortex by

means of synthetic aperture magnetometry. NeuroImage 20, 995–1005.

Howard III, M.A., Volkov, I.O., Abbas, P.J., Damasio, H., Ollendieck,

M.C., Granner, M.A., 1996. A chronic microelectrode investigation of

the tonotopic organization of human auditory cortex. Brain Res. 724,

260–264.

Howard, M.A., Volkov, I.O., Mirsky, R., Garell, P.C., Noh, M.D., Granner,

M., Damasio, H., Steinschneider, M., Reale, R.A., Hind, J.E., Brugge,

J.F., 2000. Auditory cortex on the human posterior superior temporal

gyrus. J. Comp. Neurol. 416, 79–92.

Kaas, J.H., Hackett, T.A., 2000. Subdivisions of auditory cortex and

processing streams in primates. Proc. Natl. Acad. Sci. U. S. A. 97,

11793–11799.

Kaas, J.H., Hackett, T.A., Tramo, M.J., 1999. Auditory processing in

primate cerebral cortex. Curr. Opin. Neurobiol. 9, 164–170.

Lauter, J.L., Herscovitch, P., Formby, C., Raichle, M.E., 1985. Tonotopic

organization in human auditory cortex revealed by positron emission

tomography. Hear. Res. 20, 199–205.

Lee, Y.S., Lueders, H., Dinner, D.S., Lesser, R.P., Hahn, J., Klem, G., 1984.

Recording of auditory evoked potentials in man using chronic subdural

electrodes. Brain 107, 115–131.

Lemieux, L., Leduc, A., 1992. Equivalent source estimation based on the

calculation of the electric field from depth EEG data. IEEE Trans.

Biomed. Eng. 39, 805–817.

Liegeois-Chauvel, C., Musolino, A., Chauvel, P., 1991. Localization of the

primary auditory area in man. Brain 114, 139–153.

Lockwood, A.H., Salvi, R.J., Coad, M.L., Arnold, S.A., Wack, D.S.,

Murphy, B.W., Burkard, R.F., 1999. The functional anatomy of the

normal human auditory system: responses to 0.5 and 4.0 kHz tones at

varied intensities. Cereb. Cortex 9, 65–76.

Logothetis, N.K., Pauls, J., Augath, M., Trinath, T., Oeltermann, A., 2001.

Neurophysiological investigation of the basis of the fMRI signal. Nature

412, 150–157.

B. Yvert et al. / NeuroImage 28 (2005) 140–153 153

Lutkenhoner, B., 2003. Magnetoencephalography and its Achilles’ heel.

J. Physiol. (Paris) 97, 641–658.

Lutkenhoner, B., Steinstrater, O., 1998. High-precision neuromagnetic

study of the functional organization of the human auditory cortex.

Audiol. Neuro-Otol. 3, 191–213.

Lutkenhoner, B., Krumbholz, K., Lammertmann, C., Seither-Preisler, A.,

Steinstrater, O., Patterson, R.D., 2003. Localization of primary

auditory cortex in humans by magnetoencephalography. NeuroImage

18, 58–66.

Matsuura, K., Okabe, Y., 1995. Selective minimum-norm solution of the

biomagnetic inverse problem. IEEE Trans. Biomed. Eng. 42, 608–615.

Matsuura, K., Okabe, Y., 1997. A robust reconstruction of sparse

biomagnetic sources. IEEE Trans. Biomed. Eng. 44, 720–726.

Morosan, P., Rademacher, J., Schleicher, A., Amunts, K., Schormann, T.,

Zilles, K., 2001. Human primary auditory cortex: cytoarchitectonic

subdivisions and mapping into a spatial reference system. NeuroImage

13, 684–701.

Musolino, A., Tournoux, P., Missir, O., Talairach, J., 1990. Methodology of

‘‘in vivo’’ anatomical study and stereo- electroencephalographic

exploration in brain surgery for epilepsy. J. Neuroradiol. 17, 67–102.

Pandya, D.N., 1995. Anatomy of the auditory cortex. Rev. Neurol. (Paris)

151, 486–494.

Pantev, C., Bertrand, O., Eulitz, C., Verkindt, C., Hampson, S., Schuierer,

G., Elbert, T., 1995. Specific tonotopic organizations of different

areas of the human auditory cortex revealed by simultaneous magnetic

and electric recordings. Electroencephalogr. Clin. Neurophysiol. 94,

26–40.

Pascual-Marqui, R.D., Michel, C.M., Lehmann, D., 1994. Low resolution

electromagnetic tomography: a new method for localizing electrical

activity in the brain. Int. J. Psychophysiol. 18, 49–65.

Rademacher, J., Caviness Jr., V.S., Steinmetz, H., Galaburda, A.M., 1993.

Topographical variation of the human primary cortices: implications for

neuroimaging, brain mapping, and neurobiology. Cereb. Cortex 3,

313–329.

Rademacher, J., Morosan, P., Schormann, T., Schleicher, A., Werner, C.,

Freund, H.J., Zilles, K., 2001. Probabilistic mapping and volume

measurement of human primary auditory cortex. NeuroImage 13,

669–683.

Rauschecker, J.P., 1998. Cortical processing of complex sounds. Curr.

Opin. Neurobiol. 8, 516–521.

Rivier, F., Clarke, S., 1997. Cytochrome oxidase, acetylcholinesterase, and

NADPH-diaphorase staining in human supratemporal and insular

cortex: evidence for multiple auditory areas. NeuroImage 6, 288–304.

Scheich, H., Baumgart, F., Gaschler-Markefski, B., Tegeler, C., Tempel-

mann, C., Heinze, H.J., Schindler, F., Stiller, D., 1998. Functional

magnetic resonance imaging of a human auditory cortex area

involved in foreground–background decomposition. Eur. J. Neurosci.

10, 803–809.

Scherg, M., Von Cramon, D., 1986. Evoked dipole source potentials of the

human auditory cortex. Electroencephalogr. Clin. Neurophysiol. 65,

344–360.

Schonwiesner, M., von Cramon, D.Y., Rubsamen, R., 2002. Is it tonotopy

after all? NeuroImage 17, 1144–1161.

Steinschneider, M., Volkov, I.O., Noh, M.D., Garell, P.C., Howard III,

M.A., 1999. Temporal encoding of the voice onset time phonetic

parameter by field potentials recorded directly from human auditory

cortex. J. Neurophysiol. 82, 2346–2357.

Talavage, T.M., Ledden, P.J., Benson, R.R., Rosen, B.R., Melcher, J.R.,

2000. Frequency-dependent responses exhibited by multiple regions in

human auditory cortex. Hear. Res. 150, 225–244.

Talavage, T.M., Sereno, M.I., Melcher, J.R., Ledden, P.J., Rosen, B.R.,

Dale, A.M., 2004. Tonotopic organization in human auditory cortex

revealed by progressions of frequency sensitivity. J. Neurophysiol. 91,

1282–1296.

Uutela, K., Hamalainen, M., Somersalo, E., 1999. Visualization of

magnetoencephalographic data using minimum current estimates.

NeuroImage 10, 173–180.

Wessinger, C.M., VanMeter, J., Tian, B., Van Lare, J., Pekar, J.,

Rauschecker, J.P., 2001. Hierarchical organization of the human

auditory cortex revealed by functional magnetic resonance imaging.

J. Cogn. Neurosci. 13, 1–7.

Woods, D.L., Alain, C., Covarrubias, D., Zaidel, O., 1995. Middle-latency

auditory evoked potentials to tones of different frequency. Hear. Res.

85, 69–75.

Yvert, B., Crouzeix, A., Bertrand, O., Seither-Preisler, A., Pantev, C., 2001.

Multiple supratemporal sources of magnetic and electric auditory

evoked middle latency components in humans. Cereb. Cortex 11,

411–423.

Yvert, B., Fischer, C., Guenot, M., Krolak-Salmon, P., Isnard, J., Pernier,

J., 2002. Simultaneous intracerebral EEG recordings of early

auditory thalamic and cortical activities in human. Eur. J. Neurosci.

16, 1146–1150.