combined mapping of human auditory eeg and meg responses · combined mapping of human auditory eeg...

Combined mapping of human auditory EEG and MEG responses

Minna Huotilainena,*, Istvan Winklera,b, Kimmo Alhoa,c, Carles Escerac,Juha Virtanena,d, Risto J. Ilmoniemie, Iiro P. Jaaskelainena,e,

Eero Pekkonena,e,f, Risto Naatanena

aCognitive Brain Research Unit, Department of Psychology, P.O. Box 13, FIN-00014 University of Helsinki, Helsinki, FinlandbDepartment of Psychophysiology, Institute for Psychology, Hungarian Academy of Sciences, Budapest, Hungary

cNeurodynamics Laboratory, Department of Psychiatry and Clinical Psychobiology, University of Barcelona, Barcelona, Catalonia, SpaindDepartment of Radiology, Helsinki University Central Hospital, FIN-00029 HYKS, Helsinki, Finland

eBioMag Laboratory, Medical Engineering Centre, Helsinki University Central Hospital, P.O. Box 508, FIN-00029 HYKS, Helsinki, FinlandfDepartment of Neurology, University of Helsinki, Helsinki, Finland

Accepted for publication: 22 January 1998

Abstract

Auditory electric and magnetic P50(m), N1(m) and MMN(m) responses to standard, deviant and novel sounds were studied by recordingbrain electrical activity with 25 EEG electrodes simultaneously with the corresponding magnetic signals measured with 122 MEGgradiometer coils. The sources of these responses were located on the basis of the MEG responses; all were found to be in the supra-temporal plane. The goal of the present paper was to investigate to what degree the source locations and orientations determined from themagnetic data account for the measured EEG signals. It was found that the electric P50, N1 and MMN responses can to a considerabledegree be explained by the sources of the corresponding magnetic responses. In addition, source-current components not detectable byMEG were shown to contribute to the measured EEG signals. 1998 Elsevier Science Ireland Ltd. All rights reserved

Keywords:Auditory evoked responses; Electroencephalography; Equivalent current dipole; Magnetoencephalography; P50; N1; Mismatchnegativity; Source localisation; Event-related potentials

1. Introduction

Postsynaptic neuronal currents of a large number of syn-chronously-active pyramidal cells give rise to electric andmagnetic field changes that can be measured on the scalpand outside the head. The measured electric potentials arisefrom both tangential and radial (with respect to the scalp)components of these postsynaptic currents (Regan, 1989).The corresponding magnetic field changes reflect the tan-gential component of these currents (Ha¨malainen et al.,1993; Naatanen et al., 1994). The different conductivitiesof brain tissue, skull and scalp significantly influence thescalp-recorded electric signals. In contrast, as the magnetic

permeabilities of these structures are practically equal, theinfluence on the signals recorded in magnetoencephalogra-phy (MEG) is slighter and easier to take into account.

Use of a spherical head model is more suitable for sourcelocalisation of magnetic than for electric responses (Yvert etal., 1996). However, magnetic fields produced by currentslocated close to the centre of the sphere are weak and hard todetect, requiring the use of a whole-head magnetometer andindividual, realistic head models (Tesche, 1996; Tesche etal., 1996; Tesche and Karhu, 1997). In contrast, deepsources contribute significantly to the scalp-recorded elec-tric signals (Ha¨malainen et al., 1993). Measuring the MEGand the EEG simultaneously provides one with the oppor-tunity to study activity from deep and radial sources as wellas to locate accurately the tangential component of thisbrain activity.

The auditory electric P50 response and its magnetic coun-terpart, P50m, are assumed to originate from the primary

Electroencephalography and clinical Neurophysiology 108 (1998) 370–379

0168-5597/98/$19.00 1998 Elsevier Science Ireland Ltd. All rights reservedPII S0168-5597(98)00017-3 EEP 97706

* Corresponding author. Cognitive Brain Research Unit, Department ofPsychology, P.O. Box 13, FIN-00014 University of Helsinki, Finland. Tel.:+358 9 19123451; fax: +358 9 19122924;e-mail: [email protected]

auditory cortex or from some adjacent brain area (Liegois-Chauvel et al., 1994; Ma¨kelaet al., 1994), thus enabling oneto locate the primary sensory area.

The auditory electric N1 response (Vaughan and Ritter,1970), although composed of several components (Naät-anen and Picton, 1987), has its major sources in the supra-temporal area. The magnetic counterpart of N1, the N1m,reflects the tangential part of this supratemporal N1 compo-nent. Recent results from studies with variable inter-stimu-lus intervals (ISI) suggest multiple sources for the N1m (Luët al., 1992; Sams et al., 1993; Loveless et al., 1996), but inmost cases, the N1m can be modelled by a single dipole, asthese generators are very close to each other. Ma¨kela et al.(1994) found that the N1m is located posterior to the P50m.

The mismatch negativity (MMN, Naätanen et al., 1978),and its magnetic counterpart, MMNm (Hari et al., 1984), is aresponse reflecting a pre-attentive auditory-change detec-tion process based on sensory memory traces (for a review,see Naätanen, 1992). It is elicited by infrequent auditorystimuli (deviants) in repetitive sequences of astandardsound. The main MMN generators are located in the tem-poral lobes in the left and right auditory cortices (Scherg etal., 1989; Giard et al., 1990; Halgren et al., 1995; Kropotovet al., 1995; Leva¨nen et al., 1996). An additional frontalgenerator of MMN has also been suggested (Giard et al.,1990; Alho et al., 1994; Molna´r et al., 1995; Leva¨nen et al.,1996). The MMNm (Hari et al., 1984) is generated a fewmillimetres anterior to the source of the N1m (Sams et al.,1991; Cse´pe et al., 1992; Huotilainen et al., 1993; Tiitinen etal., 1993).

The aim of the present study was to investigate the cor-respondence and differences between the electric and mag-netic P50(m), N1(m), and MMN(m) responses, bysimultaneous EEG and MEG measurements. The degreeto which sources located on the basis of the MEG recordingalone could explain the EEG data was used to determine theextent of the contribution from sources not detected byMEG to the corresponding EEG responses.

2. Methods

2.1. Subjects, stimulation, and recording

Eight healthy right-handed adults (aged 22–38 years, 6males) volunteered as subjects. None of them had a historyof hearing disorders or neurological diseases.

Standardtones were presented with a probability of 85%,and randomly-embeddeddeviant tones andnovel soundshad a probability of 7.5% each. The frequencies of the sinu-soidal standard and the deviant tones were 600 and 660 Hz,respectively. The novel sounds were selected from a groupof 60 complex ‘natural’ sounds having multiple spectralcomponents within the frequency range of 300 Hz–5 kHz.Each deviant and novel sound was preceded by at least onestandard. All stimuli were delivered at a sound pressure

level of approximately 75 dB at the earpiece. The durationof the stimuli was 200 ms, including 10 ms rise and falltimes, and the constant stimulus onset-to-onset time was800 ms.

The subjects were instructed to ignore the tones duringthe experiment. They sat in a chair with the head inside thehelmet-shaped Neuromag 122 channel MEG instrument andwatched a silent video film. The stimuli were deliveredbinaurally through plastic tubes and earpieces, using a cor-rection filter to compensate for the acoustical properties ofthe tubes (NeuroScan, USA). MEG with 122 channels (Aho-nen et al., 1993) and EEG with 25 electrodes (see Fig. 1)were recorded in a magnetically-shielded room (Euroshield,Finland). Each two channel sensor unit measures two inde-pendent magnetic field (B) gradient components,dBz/dx anddBz/dy, Bz being normal to the local scalp surface. The gold-plated electrodes were placed on the scalp according to anextended international 10–20 system and were referred toan electrode at the nose. The exact 3D locations of theelectrodes were measured using an Isotrak 3D digitiser.The recording passband was 0.03–100 Hz, the samplingrate was 397 Hz.

The position of the head with respect to the instrumentwas determined by measuring magnetic fields produced by 3marker coils on the scalp (Ahlfors and Ilmoniemi, 1989),whose locations in relation to cardinal points of the headwere determined before the experiment. The vertical andhorizontal electro-oculogram (EOG) were also recordedbipolarly from electrodes placed above and below the lefteye and from electrodes placed at the outer canthi of theright and the left eye. Epochs with an EOG change exceed-ing 150mV, an EEG change exceeding 500mV, and/or anMEG change exceeding 1.5 pT/cm were discarded, togetherwith the responses to the first few stimuli of each sequence.In addition, responses to standard stimuli following a devi-ant or a novel sound were also discarded because thesestimuli may elicit different components than the rest ofthe standards (Sams et al., 1984; Nousak et al., 1996; Wink-ler et al., 1996).

2.2. Data analysis

The data were averaged on-line using a data-collectionwindow starting at 150 ms before, and ending at 800 msafter, the beginning of the sound. Responses were filteredwith a passband of 1–30 Hz. All data were baseline-cor-rected using a prestimulus period of 50 ms. The most pro-minent responses were the P50(m) and N1(m) to thestandard tones. The deviant and novel difference curveswere calculated by subtracting the response to the standardstimulus from those to the deviant and novel stimuli, tomeasure the MMN(m) component elicited by these sounds.In the novel-minus-standard difference curve, a responseresembling the MMN(m) was observed. It will be termedthenovel-MMN(m) to distinguish it from the MMN(m) eli-cited by the deviant stimuli (termed thedeviant-MMN(m)).

371M. Huotilainen et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 370–379

The latencies of these responses were determined at Cz forthe P50 and N1, and Fz for the deviant-MMN and novel-MMN of the electric data and, for the magnetic data, at the

temporal channel pair showing the highest-amplituderesponse (determined as the square root of the squaredsum of the two field components). In all statistical compar-

Fig. 1. The magnetic and electric responses of subject S5. Left and middle: magnetic responses to standard and deviant stimuli viewed from left and right inthe helmet-shaped array of 122 channels. The horizontal (above) and vertical (below) gradient components of the magnetic field are displayed from eachlocation. Right: electric responses to standard and deviant stimuli displayed on a schematic head viewed from above, at the location of the correspondingelectrode. The nose points upwards in the figure. Top: the P50(m) and N1(m) components are marked on the responses to standard stimuli. In the inserts,enlarged views of the same responses from the selected channels (thick lines mark the locations) are shown. Bottom: the deviant-MMN(m) component isshown on the deviant-standard difference curves. In the enlarged inserts, the responses to deviants (red line) and standards (black line) from selected channels(marked by the thick lines) are presented with the deviant-MMN(m) component shaded.

372 M. Huotilainen et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 370–379

isons, analyses of variance (ANOVAs) for repeated mea-sures were used, applying the Huynh–Feldt correction whenappropriate.

2.3. Head model

Equivalent current dipole (ECD; see e.g. Ha¨malainen etal., 1993) fitting of both MEG and EEG data was done inspherical head models using Neuromag and BESA software(Neuromag, Esproo, Finland). The ECD parameters weredetermined to explain the measured data optimally in theleast-squares sense. The location and size of the sphere weredetermined individually for each subject by fitting a sphereto the individually-measured electrode locations. In EEG, a4 shell model was used; the thicknesses of the scalp, skulland cerebrospinal fluid were taken to be 6, 7 and 1 mm,respectively, their conductivities were assumed to be 0.33,0.0042 and 1Q/m, respectively, and a conductivity of 0.33Q/m was estimated for the innermost sphere, correspondingto the brain tissue. For the MEG modelling, spherical mod-els were used with the same, individually-determined cen-tres of symmetry as in EEG. The sphere origin locations andradii are presented in Table 1.

2.4. Model I: one tangential dipole in each hemisphere

The sources of the P50m, N1m and MMNm responseswere modelled with ECDs of freely varyingstrength, loca-tion and (tangential)orientation. The location and orienta-tion of the source of the magnetic field measured at aselection of 44 channels over each hemisphere was firstdetermined with one dipole at a time. Fitting of the strengthsof two simultaneously-active dipoles was done using

periods of 10, 20 and 40 ms, centred around the peak ofthe magnetic response for the P50m, N1m and MMNm,respectively, and using both 122 magnetic channels inMEG and all 25 electrodes in EEG. Theresidual variance,which indicates the amount of MEG and EEG data leftunexplained by these dipoles, was determined, and an aver-age over the 10, 20 and 40 ms periods was calculated usingall channels and all electrodes.

2.5. Model II: two tangential dipoles in each hemisphere

The orthogonaltangentialpair was added to each dipoleof Model I having an identical location with the originaldipole, separately in each hemisphere, and the sourcestrengths were determined for all channels/electrodes ofthe MEG and EEG data using these dipole pairs. ThusModel II corresponds to one dipole rotating in the tangentialplane in each hemisphere.

2.6. Model III: EEG data analysis with 3 orthogonaldipoles in each hemisphere

For the EEG data, the third orthogonal dipole (theradialone) was set up at the same locations as the dipoles ofModels I and II in each hemisphere and the source strengthswere recalculated. Model III corresponds to one dipolerotating freely in each hemisphere, each dipole being a‘regional source’ in Scherg’s terminology (Scherg, 1990).

3. Results

3.1. Peak latencies

The mean (±standard error) latencies of the electric P50and N1 responses elicited by the standard tones were 62± 6and 104± 4 ms, respectively, and the mean latencies of themagnetic P50m and N1m responses were 62± 6 and 99± 5ms in the left and 56± 7 and 96± 4 ms in the right hemi-sphere, respectively. The mean latencies of the electric devi-ant- and novel-MMN responses were 153± 6 and 111± 6ms, respectively. The corresponding magnetic responsespeaked at 143± 6 and 120± 7 ms in the left and 147± 5and 110± 8 ms in the right hemisphere. No significant dif-ferences were found between the MEG left and right hemi-sphere and EEG peak latencies for any of the components.

3.2. Source strengths

The source strengths were determined as averages of theModel I dipole moments over the 10, 20 and 40 ms latencyranges of the P50m, N1m and MMNm components, respec-tively, using data from all 122 MEG channels. Because ofthe small responses to the standard tones, the ECDs were notdeterminable in the left hemisphere for the P50m and N1mresponses in subjects 1 and 2.

Table 1

Location and radius of the outermost sphere of the volume model for EEGdata determined on the basis of the individually-measured electrode loca-tions

Spherical head model

Subject Radius of theoutermost sphere(mm)

Sphere origin (mm)

x y z

S1 92 −0.8 19.4 35.9S2 91 −1.5 12.0 39.9S3 93 0.5 1.5 35.4S4 93 0.3 8.4 38.9S5 90 0.4 5.5 46.0S6 95 4.2 7.3 48.8S7 89 2.0 6.9 41.6S8 89 −0.4 11.8 41.9

A 4 shell model for EEG data analysis was constructed on the basis of thisoutermost sphere. In MEG, a one shell spherical volume model was used,whose origin was the same as in EEG. The positivex-axis penetratesthrough the right preauricular point, the positivey-axis through the nasionand the positivez-axis approximately through vertex.


Significant inter-hemispheric differences were found inthe response strengths. Single-factor ANOVAs indicatedhigher dipole moments in the left than in the right hemi-sphere for the N1m (n = 7, mean difference 1 nAm,P ,0.01) and deviant-MMNm (n = 8, mean difference 15 nAm,P , 0.01). A similar tendency was found for the P50m(n = 6, mean difference 4 nAm,P , 0.07).

3.3. Source locations

As seen from Figs. 2 and 3 and Table 2, the co-ordinatesfor the P50m, N1m, deviant-MMNm and novel-MMNmECDs suggest that the generators of these responses arelocated in the auditory cortex in the superior temporalplane. The location of the N1m response was consideredto be a landmark against which other source locationswere compared.

In the right hemisphere, the ECD locations of the P50m,deviant-MMNm, and novel-MMNm were, on average, ante-rior to that of the N1m by 11, 18 and 5 mm, respectively, butthese differences did not reach statistical significance. In theleft hemisphere, the ECD location of the deviant-MMN wassignificantly medial and inferior to the N1m location (n = 8,mean differences 8 and 12 mm,P , 0.05 andP , 0.01,respectively).

Although the inter-hemispheric differences of the dipolelocations in the anterior-posterior direction (see Table 2) did

not reach statistical significance, the tendencies corroboratethe findings of previous anatomical (Gershwind andLevitsky, 1968) and functional MEG studies (Ma¨kela etal., 1994) demonstrating the more anterior location ofright-hemisphere auditory cortex compared with that ofthe left hemisphere.

3.4. MEG residual variances

The mean residual variances of the MEG models arepresented in Table 3. The magnetic P50m and N1mresponses could be explained with reasonable confidenceby only one (tangential) dipole in each hemisphere. Theresidual variances over a period of 10 and 20 ms aroundthe peak of the P50m and N1m responses were both 15.3%when the data from all 122 magnetic channels was takeninto account in Model I (one dipole in each hemisphere).Adding the orthogonal tangential component to the solution(switching from Model I to Model II) reduced the residualvariances of the P50m and N1m responses, to 14.6 and14.0%, respectively (by 5% and 9% of the Model I residualvariance).

For the deviant- and novel-MMNm responses, the resi-dual variances over a period of 40 ms around the peak of theresponse in Model I were 27.2% and 21.7%. When Model IIwas applied, the variances of the deviant- and novel-MMNm responses were reduced to 24.0% and 20.4%,

Fig. 2. The MEG-based ECDs for subject S5. The dipole locations and orientations were fitted on the basis of the magnetic responses, shown here for subjectS5, superimposed on his MR images. The MR images were obtained with a 1.5 T Siemens Vision system of the Department of Radiology, HelsinkiUniversity Central Hospital. An MP-RAGE imaging sequence was used with 1 mm slice thickness and an 1× 1 mm in-plane pixel size. The dipole locationsand orientations were determined separately for each hemisphere using information from 44 magnetic channels centred around the auditory cortex in eachhemisphere. The orthogonal tangential dipole was added in Model II (two dipoles in each hemisphere). The original dipole orientation (blue and red for N1mand MMNm, respectively) and its orthogonal tangential counterpart (lighter blue and lighter red) are displayed on the 3 views (two parasagittal and onecoronal) of the MR images of S5 (from the left, back and right). Left: the temporal behaviour of the left-hemisphere sources (source strength as a function oftime) calculated using information from all 122 channels. On the top, the temporal behaviour of the N1m dipole from Model I (one dipole in eachhemisphere) is shown in blue. Below, the temporal behaviour of the N1m ECDs from Model II are shown. On the bottom, the temporal behaviour ofthe MMNm ECDs (red) from Models I and II are presented. Right: the temporal behaviour of the right-hemisphere ECDs are presented similarly to the lefthemisphere. The residual variance of N1m and deviant-MMNm ECDs of Models I and II are presented with a green line. This value is the amount of data leftunexplained with these dipoles, smaller values denote better explanation. The models are optimised for 20 and 40 ms periods around the peak of the N1(m)and deviant-MMN(m) responses, respectively.


respectively (by 12% and 6% of the Model I residual var-iances).

3.5. EEG residual variances

The mean residual variances of the EEG models are pre-sented in Table 4. The mean residual variances of the P50and N1 responses in Model I were 21.2% and 19.5%,respectively. When Model II was applied, the residual var-iances were reduced to 16.8% and 13.8%, respectively (by21% and 29% of the Model I residual variances). Finally,when Model III was applied, the residual variances werefurther reduced to 6.6% and 7.9%, respectively (by 61%and 43% of the Model II variances).

The mean residual variances of the deviant-MMN and

novel-MMN responses in Model I were 21.0 and 37.8%,respectively. When Model II was applied, the residual var-iances were reduced to 13.5% and 18.1% (by 36% and 52%of the Model I residual variances), and with Model III, afurther reduction to 6.6% and 9.3% (51% and 49% of theModel II variances) was observed.

To assess the quality of the MEG-based models fordescribing the EEG responses (with respect to the qualityof the original MEG model), the ratio between the resi-dual variances in the EEG and MEG was calculated sepa-rately in Models I and II for each target component ineach subject. The smaller this number, the better theMEG-based model accounts for the EEG data. A reduc-tion of this number between Models I and II would meanthat the additional tangential component improved (in

Fig. 3. Temporal behaviour of the electrical sources. The dipole locations and orientations were taken from MEG Models I and II (colours correspond tothoseof ), the radial dipole was added for Model III (colours: light blue and light red). The ECD source potentials were calculated by the BESA software (Schergand Berg, 1992) using information from all 25 electrodes. Left: the temporal behaviour of the left-hemisphere ECDs in the electric data. On the top panel, thesource potentials from the N1 Models I, II and III are presented separately. On the bottom panel, source potentials obtained in Models I, II and III for thedeviant-MMN are presented. Right: temporal behaviour of sources in the right hemisphere. On the extreme right, the residual variance curves (green colour)for each model/component are shown.


terms of the unaccounted-for variance) the description ofthe EEG response to a larger degree than the descriptionof the MEG data. The ratios between the EEG and MEGresidual variances for the P50(m), N1(m), deviant-MMN(m) and novel-MMN(m) in Model I were, on aver-age, 1.39, 1.27, 0.77 and 1.74, respectively. In ModelII, the corresponding ratios were 1.15, 0.99, 0.55 and0.89.

The difference between the Model I and the Model IIEEG/MEG residual variance ratios was tested by single-factor ANOVAs, separately for each component. All ofthese tests, except the one for the deviant-MMN(m), pro-duced significant results (P50(m):F(1,5) = 9.74,P , 0.05;N1(m): F(1,6) = 8.69,P , 0.05; deviant-MMN(m):F(1,7)= 3.16, not significant; novel-MMN(m):F(1,6) = 28.55,P , 0.01). This means that by adding the orthogonal tan-gential dipole, the MEG-based models of P50(m), N1(m)and the novel-MMN(m) improved the description of theEEG data significantly more than that of the MEGresponses. To assess the relative quality of the modelsacross different components, a two-factor ANOVA (com-ponent: N1(m), deviant-MMN(m), novel-MMN(m) timesModel I, II) was calculated. All measures were availableonly for 6 subjects. The P50(m) component was omittedfrom this analysis because the lack of acceptable ECDs intwo subjects for this component (see above) would have

further reduced the number of subjects included in the testto 4. The significant interaction between the two factors(F(2,10) = 5.06,P , 0.05, Huynh–Feldte = 0.99) resultedfrom a much larger EEG/MEG residual variance ratioreduction for the novel-MMN(m) than for any other com-ponent (F(1,10) = 17.37,P , 0.01 for the novel-MMN(m)vs. F(1,10) = 4.49 andF(1,10) = 1.78, for the N1(m) anddeviant-MMN(m) components, respectively; post-hoc testswere conducted by Scheffe´-type pairwise comparisons).The significant result of the Component factor (F(2,10) =5.13,P , 0.05, Huynh–Feldte = 0.78) was due to the largedifference between the residual variance ratios of the novel-and the deviant-MMN(m) (F(2,10) = 5.23, P , 0.05). Inaccordance with the single-factor ANOVA results, theModel factor showed a significant reduction of the EEG/MEG residual variance ratios from Model I to Model II(F(1,5) = 13.02,P , 0.05).

4. Discussion

4.1. Source locations

The locations of the P50m, N1m, and MMNm responsesare in line with the findings of previous studies indicatingthat supratemporal generators are the main sources of theseresponses (Hari, 1990; Pantev et al., 1990, 1995; Rogers etal., 1990; Makelaet al., 1994; Reite et al., 1994; Nakasato etal., 1994). That deviant-MMNm and P50m source locationsare anterior to N1m is also in agreement with recent findings(Sams et al., 1991; Cse´pe et al., 1992; Huotilainen et al.,1993; Levanen et al., 1993; Tiitinen et al., 1993; Alho et al.,1996), although some of these differences did not reachsignificance in the present data.

4.2. Peak latencies

No significant differences were found for the peaklatency of any of the responses when the latencies fromFz/Cz of the EEG and latencies from the left and righthemispheres in MEG were compared. Thus we can assumethat the responses in MEG and EEG at least partly representthe same neuronal sources.

The latency of the novel-MMN(m) response was clearlyearlier than that of the deviant-MMN(m). This may be due

Table 2

Across-subject mean locations, orientations and strengths of the P50(m),N1(m), deviant-MMN(m) and novel-MMN(m) ECDs

Results of dipole modelling: dipole parameters

Location (mm) Strength(nAm)

Orientation(°)

x y z Q a

P50m left −46 5 51 8 34P50m right 58 17 53 4 47N1m left −48 8 57 6 −113N1m right 55 6 55 5 −120Deviant-MMNm left −40 11 45 26 −136Deviant-MMNm right 51 24 52 11 −141Novel-MMNm left −46 3 53 23 −145Novel-MMNm right 49 11 54 26 −129

The anglea is determined counterclockwise (viewed from the right) withthe positivey-axis. The ECD strengths are from MEG Model I.

Table 3

Across-subject mean MEG residual variances (amount of data left unex-plained by the model) of the P50m, N1m, deviant-MMNm and novel-MMNm for the 3 different models

Residual variances for 122 channel MEG

Model P50m N1m Deviant-MMNm

Novel-MMNm

I 15.3 15.3 27.2 21.7II 14.6 14.0 24.0 20.4

Table 4

Across-subject mean EEG residual variances (amount of data left unex-plained by the model) of the P50, N1, deviant-MMN and novel-MMN forthe 3 different models

Residual variances for 25 channel EEG

Model P50 N1 Deviant-MMN Novel-MMN

I 21.2 19.5 21.0 37.8II 16.8 13.8 13.4 18.1III 6.6 7.9 6.6 9.3


to the fact that the novel-MMN(m) response was only partlycaused by a true MMN-type response. A simultaneousenhanced N1(m) was probably overlapping this response(Scherg et al., 1989). The N1(m) to novel sounds was pre-sumably enhanced, due to their wider frequency-spectraactivating such frequency-specific neurons as had notbecome refractory by the repetition of the 600 Hz standardtone.

4.3. MEG residual variances

Model I dipoles account well for the P50m and N1mresponses. The larger deviant-MMNm and novel-MMNresidual variances can be explained partly by the fact thatthese components are obtained by subtraction (whichincreases noise level), and partly by the fact that the deviantand novel responses had originally clearly fewer summa-tions in the average than the standard responses.

The transition from Model I to Model II reduced theresidual variations by only 5–12% of the original residualvariance. This reduction is so small that one can concludethat Model I (one dipole in each hemisphere) is sufficient fordescribing the magnetic P50m, N1m, deviant-MMNm andnovel-MMNm responses. In general, no consistent peaklatency differences were found across subjects betweenthe two source potential curves of Model II within the mod-elled intervals. This suggests that the orthogonal tangentialdipole accounted for temporal changes in the source direc-tion rather than describing additional sources. Model II ofthe deviant-MMN(m) in the right hemisphere (Fig. 2) is anexception. The small early (approx. 80–120 ms) wave of theorthogonal tangential source potential might reflect the con-tribution from an increased N1(m) component (cf. Scherg etal., 1989).

4.4. EEG residual variances

The lower EEG/MEG residual variance ratios in Model IIthan in Model I suggest that addition of the second dipolereduces the residual variance significantly more in EEG thanin MEG. This indicates that the MEG-based dipole locationsare not optimal to explain even the tangential part of theEEG responses. This may be due to contributions from deepsources in the EEG data.

The MEG-based models are not equally good in describ-ing the different EEG components. Results of the statisticalanalysis (specifically the interaction between the Compo-nent and Model factors) suggest that the source locationcalculated for the novel-MMN component from the MEGdata was less optimal than that for the other components.This might be due to the novel-MMN reflecting two sepa-rate processes (and, therefore, probably deriving from mul-tiple sources). As was mentioned above, the response torare, widely-deviating sounds activates both the N1 andMMN generators (Scherg et al., 1989). It is possible thatin addition to the supratemporal N1, other subcomponents

of the N1 (Naatanen and Picton, 1987) were also differen-tially affected by the standard and the novel sounds.Attempting to describe the overlapping effects of severalgenerators by a single ECD results in relatively largeamounts of residual variance, which is then significantlyreduced by including additional dipoles in the model.

The radial component placed in the MEG-based location(replacing Model II with Model III) had a marked effect: theresidual variance was reduced by 43–61%. A more optimalsolution for EEG modelling would be to separately search alocation for the radial dipole, since the MEG-based locationmay not be optimal for the EEG data. This would enable theradial dipole to better account for the deep and radialsources not visible in MEG.

5. Conclusions

When tangential current sources are located close to thesurface of the brain, MEG-based dipole modelling is a finetool for locating the sources. The relatively high residualvariances when EEG was explained by the MEG-basedmodels show that it would be beneficial to use MEG andEEG simultaneously to aid modelling of electromagneticdata. MEG-based modelling is a good starting point fromwhich the model can be refined by adding further dipoles toaccount for the activity not detected by MEG.

The aim of the present investigation was to assess howwell MEG-based dipole models of the target componentsaccount for the corresponding electrical activity observed inEEG. In order to make MEG and EEG source modellingcomparable, only single-dipole models restricted to specificcomponents were considered. This was necessary becausethe magnetic activity measured by MEG and the electricalsignals recorded in parallel by EEG do not fully correspondto each other and, therefore, comparisons between multi-dipole models describing several components in MEG andEEG would have yielded highly complex results. In thesearch for optimal solutions, the present ECD models canbe further refined. For example, whereas the addition of anorthogonal tangential dipole at the same location (Model II)did not significantly decrease the residual variance in theMEG model, the addition of a second dipole at a differentlocation could have. Therefore, the modelling strategyapplied in the present study reflected the goals of this inves-tigation, not limitations of MEG or EEG modelling ingeneral.

Acknowledgements

We thank Ms. Suvi Heikkila¨ for her assistance in themeasurements and Mr. Harri Cole´rus for helping with thefigures. This research was supported by the Academy ofFinland, the Finnish Cultural Foundation, the FinnishFoundation for Alcohol Studies, the National Scientific


Research Fund of Hungary (OTKA T022681) and the Span-ish Ministry of Education and Culture (DGES-UE96–0038).

References

Ahlfors, S. and Ilmoniemi, R.J. Magnetometer position indicator for multi-channel MEG. In: S.J. Williamson, M. Hoke and G.L. Romani (Eds.),Advances in Biomagnetism. Plenum Press, New York, 1989, pp. 673–676.

Ahonen, A.I., Hamalainen, M.S., Kajola, M.J., Knuutila, J.E.T., Laine, P.,Lounasmaa, O.V., Simola, J.T.A. and Tesche, C.D. A 122-channelSQUID instrument for investigating the magnetic signals from thehuman brain. Phys. Scr., 1993, T49: 198–205.

Alho, K., Tervaniemi, M., Huotilainen, M., Lavikainen, J., Tiitinen, H.,Ilmoniemi, R.J., Knuutila, J. and Naätanen, R. Processing of complexsounds in the human auditory cortex as revealed by magnetic brainresponses. Psychophysiology, 1996, 33: 369–375.

Alho, K., Woods, D., Algazi, A., Knight, R.T. and Naätanen, R. Lesions offrontal cortex diminish the auditory mismatch negativity. Electroen-ceph. clin. Neurophysiol., 1994, 91: 353–362.

Csepe, V., Pantev, C., Hoke, M., Hampson, S. and Ross, B. Evoked mag-netic responses of the human auditory cortex to minor pitch changes:localization of the mismatch field. Electroenceph. clin. Neurophysiol.,1992, 84: 538–548.

Gershwind, N. and Levitsky, W. Human brain: left-right asymmetries intemporal speech region. Science, 1968, 161: 186–187.

Giard, M.-H., Perrin, F., Pernier, J. and Bouchet, P. Brain generatorsimplicated in the processing of auditory stimulus deviance: a topo-graphic event-related potential study. Psychophysiology, 1990, 27:627–640.

Halgren, E., Baudena, P., Clarke, J.M., Heit, G., Liegois, C., Chauvel, P.and Musolino, A. Intracerebral potentials to rare target and distractorauditory and visual stimuli. I. Superior temporal plane and parietal lobe.Electroenceph. clin. Neurophysiol., 1995, 94: 191–220.

Hamalainen, M., Hari, R., Ilmoniemi, R.J., Knuutila, J. and Lounasmaa,O.V. Magnetoencephalography – theory, instrumentation and applica-tions to noninvasive studies of the working human brain. Rev. Mod.Phys., 1993, 65: 413–497.

Hari, R. Magnetic evoked fields of the human brain: basic principles andapplications. Electroenceph. clin. Neurophysiol., 1990, 41: 3–12.

Hari, R., Hamalainen, M.S., Ilmoniemi, R.J., Kaukoranta, E., Reinikainen,K., Salminen, J., Alho, K., Naätanen, R. and Sams, M. Responses of theprimary auditory cortex to pitch changes in a sequence of tone pips:neuromagnetic recordings in man. Neurosci. Lett., 1984, 50: 127–132.

Huotilainen, M., Ilmoniemi, R.J., Lavikainen, J., Tiitinen, H., Alho, K.,Sinkkonen, J. and Naätanen, R. Interaction between representations ofdifferent features of auditory sensory memory. NeuroReport, 1993, 4:1279–1281.

Kropotov, J.D., Naätanen, R., Sevostianov, A.V., Alho, K., Reinikainen,K. and Kropotova, O.V. Mismatch negativity to auditory stimuluschange recorded directly from the human temporal cortex. Psychophy-siology, 1995, 32: 418–422.

Levanen, S., Ahonen, A., Hari, R., McEvoy, L. and Sams, M. Deviantauditory stimuli activate human left and right auditory cortexdifferently. Cereb. Cortex, 1996, 6: 288–296.

Levanen, S., Hari, R., McEvoy, L. and Sams, M. Responses of the humanauditory cortex to changes in one versus two stimulus features. Exp.Brain Res., 1993, 97: 177–183.

Liegois-Chauvel, C., Musolino, A., Badier, J.M., Marquis, P. and Chauvel,P. Evoked potentials recorded from the auditory cortex in man: evalua-tion and topography of the middle latency components. Electroenceph.clin. Neurophysiol., 1994, 92: 204–214.

Loveless, N., Leva¨nen, S., Jousma¨ki, V., Sams, M. and Hari, R. Temporal

integration in auditory sensory memory: neuromagnetic evidence. Elec-troenceph. clin. Neurophysiol., 1996, 100: 220–228.

Lu, Z.L., Williamson, S.J. and Kaufman, L. Human auditory primary andassociation cortex have differing lifetimes for activation traces. BrainRes., 1992, 572: 236–241.

Makela, J.P., Ha¨malainen, M., Hari, R. and McEvoy, L. Whole-head map-ping of middle-latency auditory evoked magnetic fields. Electroenceph.clin. Neurophysiol., 1994, 92: 405–414.

Molnar, M., Skinner, J.E., Cse´pe, V., Winkler, I. and Karmos, G. Correla-tion dimension changes accompanying the occurrence of the mismatchnegativity and the P3 event-related potential component. Electroenceph.clin. Neurophysiol., 1995, 95: 118–126.

Naatanen, R. Attention and Brain Function. Lawrence Erlbaum, Hillsdale,NJ, 1992.

Naatanen, R., Ilmoniemi, R.J. and Alho, K. Magnetoencephalography instudies of human cognitive brain function. Trends Neurosci., 1994, 17:389–396.

Naatanen, R., Gaillaird, A.W.K. and Ma`ntysalo, S. Early selective-atten-tion effect reinterpreted. Acta Psychologica, 1978, 42: 313–329.

Naatanen, R. and Picton, T. The N1 wave of the human electric andmagnetic response to sound: a review and an analysis of the componentstructure. Psychophysiology, 1987, 24: 375–425.

Nakasato, N., Fujita, S., Seki, K., Kawamura, T., Matani, A., Tamura, I.and Fujiwara, S. Functional localization of bilateral auditory corticesusing an MRI-linked whole head magnetoencephalography (MEG)system. Electroenceph. clin. Neurophysiol., 1994, 94: 183–190.

Nousak, J.M.K., Deacon, D., Ritter, W. and Vaughan, H.G. Jr. Storage ofinformation in transient auditory memory. Cogn. Brain Res., 1996, 4:305–317.

Pantev, C., Bertrand, O., Eulitz, C., Verkindt, C., Hapson, S., Schuierer, G.and Elbert, T. Specific tonotopic organizations of different areas ofthe human auditory cortex revealed by simultaneous magnetic andelectric recordings. Electroenceph. clin. Neurophysiol., 1995, 94:26–40.

Pantev, C., Hoke, M., Lehnertz, K., Fahrendorf, B.L.G. and Stober, U.Identification of sources of brain neuronal activity with high spatiotem-poral resolution through combination of neuromagnetic source localiza-tion (NMSL) and magnetic resonance imaging (MRI). Electroenceph.clin. Neurophysiol., 1990, 75: 173–184.

Regan, D. Human Brain Electrophysiology: Evoked Potentials and EvokedMagnetic Fields in Science and Medicine. Chapman and Hall, London,1989.

Reite, M., Sheeder, J., Teale, P., Richardson, D., Adams, M. and Simon, J.MEG based brain laterality: sex differences in normal adults. Neurop-sychologia, 1994, 33: 1607–1616.

Rogers, R.L., Papanicolaou, A.C., Baumann, S.B., Saydjari, C. andEisenberg, H.M. Neuromagnetic evidence of a dynamic excitation pat-tern generating the N100 auditory response. Electroenceph. clin.Neurophysiol., 1990, 77: 237–240.

Sams, M., Alho, K. and Naätanen, R. Short-term habituation and dish-abituation of the mismatch negativity of the ERP. Psychophysiology,1984, 21: 434–441.

Sams, M., Ha¨malainen, M., Hari, R. and McEvoy, L. Human auditorycortical mechanisms of sound lateralization: I. Interaural time differ-ences within sound. Hear. Res.,, 1993, 67: 89–97.

Sams, M., Kaukoranta, E., Ha¨malainen, M. and Naätanen, R. Corticalactivity elicited by changes in auditory stimuli: different sources forthe magnetic N100m and mismatch responses. Psychophysiology,1991, 28: 21–29.

Scherg, M. Fundamentals of dipole source potential analysis. In: F. Grand-ori, M. Hoke and G.L. Romani (Eds.), Auditory Evoked MagneticFields and Electric Potentials. Advances in Audiology, Vol. 6. Karger,Basel, 1990, pp. 40–69.

Scherg, M. and Berg, P. BESA Brain Electric Source Analysis. Version1.9, Handbook 2. NeuroScan, Herndon, Virginia, 1992.

Scherg, M., Vajsar, J. and Picton, T. A source analysis of the late humanauditory evoked potentials. J. Cogn. Neurosci., 1989, 1: 336–355.


Tesche, C.D. Non-invasive imaging of neuronal population dynamics inhuman thalamus. Brain Res., 1996, 729: 253–258.

Tesche, C.D. and Karhu, J. Somatosensory evoked magnetic fields arisingfrom sources in the human cerebellum. Brain Res., 1997, 744: 23–31.

Tesche, C.D., Karhu, J. and Tissari, S.O. Non-invasive detection of neu-ronal population activity in human hippocampus. Cogn. Brain Res.,1996, 4: 39–47.

Tiitinen, H., Alho, K., Huotilainen, M., Ilmoniemi, R.J., Simola, J. andNaatanen, R. Tonotopic auditory cortex and the magnetoencephalo-graphic (MEG) equivalent of the mismatch negativity. Psychophysiol-ogy, 1993, 30: 537–540.

Vaughan, H.G. Jr. and Ritter, W. The sources of auditory evoked responsesrecorded from the human scalp. Electroenceph. clin. Neurophysiol.,1970, 28: 360–367.

Winkler, I., Karmos, G. and Na¨atanen, R. Adaptive modeling of the unat-tended acoustic environment reflected in the mismatch negativity event-related potential. Brain Res., 1996, 742: 239–252.

Yvert, B., Bertrand, O., Echallier, J.F. and Pernier, J. Improved dipolelocalization using local mesh refinement of realistic head geometries:an EEG simulation study. Electroenceph. clin. Neurophysiol., 1996, 99:79–89.


combined mapping of human auditory eeg and meg responses · combined mapping of human auditory eeg...

Documents