Journal of Neuroscience Methods 229 (2014) 2832

Contents lists available at ScienceDirect

Journal of Neuroscience Methods

journal homepage: www.elsevier.com/locate/jneumeth

Clinical Neuroscience

A spatial approach of magnitude-squared coherence applied to

selective attention detection
Leonardo Bonato Felix a , Fernando de Souza Ranaudo b , Aluizio Daffonseca Netto b ,
Antonio Mauricio Ferreira Leite Miranda de S b,
a
b

NIAS, Departamento de Engenharia Eltrica, Universidade Federal de Vicosa, Vicosa, Minas Gerais, Brazil
Programa de Engenharia Biomdica COPPE Universidade Federal do Rio de Janeiro, Rio de Janeiro, Rio de janeiro, Brazil

h i g h l i g h t s

We acquired the EEG during an auditory selective attention protocol.

We developed a new index to detect the focus of auditory attention.
This index takes into account both the electrode location and coherence value.
The index seemed to be suitable for BCI applications.

a r t i c l e

i n f o

Article history:
Received 14 November 2013
Received in revised form 28 February 2014
Accepted 26 March 2014
Keywords:
Selective attention
Spatial Coherence
Auditory steady-state response

a b s t r a c t
Background: Auditory selective attention is the human ability of actively focusing in a certain sound
stimulus while avoiding all other ones. This ability can be used, for example, in behavioral studies and
brain-machine interface.
New method: In this work we developed an objective method called Spatial Coherence to detect the
side where a subject is focusing attention to. This method takes into consideration the Magnitude Squared
Coherence and the topographic distribution of responses among electroencephalogram electrodes. The
individuals were stimulated with amplitude-modulated tones binaurally and were oriented to focus
attention to only one of the stimuli.
Results: The results indicate a contralateral modulation of ASSR in the attention condition and are in
agreement with prior studies. Furthermore, the best combination of electrodes led to a hit rate of 82% for
5.03 commands per minute.
Comparison with existing method(s): Using a similar paradigm, in a recent work, a maximum hit rate of
84.33% was achieved, but with a greater a classication time (20 s, i.e. 3 commands per minute).
Conclusions: It seems that Spatial Coherence is a useful technique for detecting focus of auditory selective
attention.
2014 Elsevier B.V. All rights reserved.

1. Introduction
Whenever an external stimulation is applied to the sensory
system, responses are expected to occur. The stimuli, which may
be auditory (clicks), visual (a stroboscopic light) or somatosensory (electrical shocks), often exhibit a periodic pattern and hence
the responses are expected to be phase-locked to them. These

Corresponding author at: Programa de Engenharia Biomdica Coppe/UFRJ,

Universidade do Rio de Janeiro, Caixa Postal (P.O. Box) 68510, CEP (ZIP code) 21941972, Brazil. Tel.: +55 21 2562 8629; fax: +55 21 2562 8591.
E-mail address: amms@peb.ufrj.br (A.M. Ferreira Leite Miranda de S).
http://dx.doi.org/10.1016/j.jneumeth.2014.03.014
0165-0270/ 2014 Elsevier B.V. All rights reserved.

responses are called evoked responses (ER) and can be recorded

immersed in the electroencephalogram (EEG). The ER usually
presents low amplitude, in comparison with the spontaneous cerebral activity (Perry, 1966). Due to this low signal-to-noise ratio
(SNR), signal-processing techniques are often applied to turn the
ER evident in the background EEG. ERs have been widely used
in studies searching for a better understanding of several sensory modality inputs that integrate the human body (Chiappa,
1997). The recording of these responses by electroencephalography associated with application of signal processing techniques
has supplied important information on the way external stimuli
are conducted, noticed and interpreted by the central nervous
system.

L. Bonato Felix et al. / Journal of Neuroscience Methods 229 (2014) 2832

If a sound stimulus is presented in a sufciently rapid rate, Auditory Steady State Responses (ASSR) can be registered from the
human scalp. Such responses can be evoked using an amplitude
modulation signal (AM tone), applied to the human ear as stimulus.
The ASSR evoked by AM tone is characterized by an energy increase
in the modulating frequency of the power spectrum of the EEG signal (Felix et al., 2005). In comparison with transitory stimuli, AM
tone has the advantage of being closer to sounds that may happen in
nature, turning the evoked ASSRs more suitable for daily situations
(Joris et al., 2004). Most studies with ASSRs were conducted with
stimuli at 40 Hz or in the frequency range between 80 and 100 Hz
(Cohen et al., 1991; Levi et al., 1993; Ross et al., 2003). Both the
response evoked by a 40 Hz AM tone (Kuwada et al., 1986) and an
80 Hz tone (Rickards et al., 1994) are important for the evaluation
of hearing integrity, allowing the application of objective audiometry. The 40 Hz responses can also be used to monitor conscience
during general anesthesia (Plourde and Picton, 1990).
Auditory selective attention (ASA) refers to the mental capacity
of resisting to distracting stimuli and focusing on relevant information of a determined stimulus (or message), as illustrated in
the Cocktail Party Effect, which describes the skill of focusing
attention to a determined announcer in an environment in which
a mixture of conversations and background noises are observed
(Bronkhorst, 2000). ASA has been investigated in order to elucidate
the mechanism underlying some features of hearing information
control, such as the ability of selecting the relevant stimuli to
which the subject must respond appropriately (Giard et al., 2000).
By applying ASSR in ASA studies, Wolpaw et al. (2002) found an
increase in the evoked responses during selective attention. However, the effects of attention on ASSR are still a matter of debate.
Linden et al. (1987) observed no signicant changes in ASSR during ASA. In a more recent work with magnetoencephalographic
signals, Ross and collaborators (2004) observed an increase in the
40 Hz ASSR for the subjects focusing on a visual stimulus compared
to the rest condition. In addition, topographical analysis of evoked
hearing responses could provide information of cerebral structures
involved in the neurophysiological attention process (Giard et al.,
2000).
The ASSRs present lower amplitude levels in comparison with
the background EEG, which hampers the identication of the
response by means of visual inspection. In this context, objective
response detection techniques (ORD) (Dobie and Wilson, 1993)
may be used to turn the response identication process automatic, as well as to allow comparing such responses (Miranda de
S and Felix, 2002). The detection procedure is carried out by a
hypothesis test, which compares the detector estimate with a statistical threshold the critical value that is based on the detector
sampling distribution under null hypothesis (H0 ) of absence of
responses. The main advantage of this technique lies on the fact
that detectors are robust, since the critical value is not dependent
on the response waveform and neither on the SNR. Furthermore,
the probability of erroneously detecting a response (probability of
a false positive) is constant and equal to the signicance level of the
test (Dobie and Wilson, 1989; Felix et al., 2005).
The magnitude squared coherence (MSC) estimate between a
periodic stimulus and EEG was proposed by Dobie and Wilson
(1989) as an ORD method for evoked responses detection. The
greatest advantage of this technique lies on the fact that MSC can be
estimated using only the biological signal, which leads to a simpler
expression. MSC has been used to investigate visual (Miranda de
S and Felix, 2002), somatosensory (Melges et al., 2008) and auditory (Felix et al., 2005) ERs in order to detect the evoked responses
embedded in the EEG based on statistical thresholds.
A braincomputer interface (BCI) is a direct, non-cephalic communication pathway between the brain and an external device. In
this context, ASA may be used together with an ORD-technique,

29

such as the MSC, to build a communication channel between disabled people and the computer. In such an application, the subject
would actively select between two auditory stimuli and the respective ER would be detected by the ORD technique. The development
of BCI systems based on ASA has the main advantage of allowing
the application in severally incapacitated patients, who present difculties in controlling the ocular movement, since these systems
are vision free (Kim et al., 2011). The combined use of selective
attention power and ASSR as a tool for interpreting cognitive activity becomes promising and the appropriate classication of these
responses according to attention status supply a paradigm for the
BCI hearing system. Additionally, recent studies reported lateralization evidences, i.e. greater amplitude of the ASSR in the condition
of attention for a contralateral stimulus (Ross et al., 2004; Gander
et al., 2007; Muller et al., 2009). This should indicate that an ASA
detection technique must also consider the spatial component, that
is, the distribution of responses in electrodes placed in the scalp.
Thus, this paper proposes a method for detection/classication
of ASA evoked by ASSR. This method takes into consideration the
ORD technique MSC of each EEG channel and the topographic distribution of responses among electrodes.
2. Methods
2.1. Experimental protocol
2.1.1. Casuistics
EEG signals were collected from fourteen adult men aged
between 19 and 28 years (22.4 2.3 years) with no symptoms of
neurological pathologies or hearing loss, which had been veried
by means of conventional audiometry. The experimental protocol
was approved by the local ethics committee.
2.1.2. Data acquisition
EEG signals were collected with a 36-channel biological signal
amplier (BrainNET, BNT-36, EMSA, Brazil) with 0.1 Hz and 100 Hz
band-pass lter. Sampling frequency of 600 Hz and 60 Hz notch lter were used. The EEG signals were registered from non-invasive
electrodes placed on the scalp, according to the international 1020
system, with reference to electrode Oz in order to obtain higher
SNR for ASSR (Van Dun and Wouters, 2009). Silver/silver chloride (Ag/AgCl) electrodes with 10 mm diameter were used and
impedance was always maintained inferior to 5 k.
2.1.3. Auditory stimulation
AM tones were used for stimulation, with 32 and 38 Hz modulation and 500 and 2000 Hz carrier frequencies, on the left and
right ears, respectively. The modulation depth was 100%. In order
to avoid spectral leakage, the modulation frequencies were corrected in such a way that each window with 1024 points contained a
prime number of cycles of the stimulating signal (Felix et al., 2005).
The stimuli were digitally produced in a personal computer, with
a sampling frequency of 44,100 Hz. The maximum amplitude was
adjusted so that the intensity of each AM tone was 75 dB SPL.1 All
stimulation experiments were conducted with the volunteer in the
interior of an audiometric cabin, with acoustic isolation, to avoid
interference of surrounding noises during the procedure. In this
work the subjects were tested only once.
The stimulation protocol was composed by the following three
steps: rst, ignoring both stimuli; second, focusing on the left stimulus only and; third, focusing on the right stimulus only. Each

1
dBSPLstandard unity for measuring sound pressure level, dened as ten times
the base 10 logarithm of the ratio of the square of observed pressure to the square
of a reference pressure level stipulated at 20 Pa.

30

L. Bonato Felix et al. / Journal of Neuroscience Methods 229 (2014) 2832

Fig. 1. Timeline to the auditory stimulus. Stimulation of the left ear (top) and right ear (bottom), with carrier frequencies of 500 Hz modulated at 32 Hz and 2000 Hz modulated
at 38 Hz, respectively.

stimulation stage lasted 2 min with 30-s intervals between stimulations. Before each stage in which the individual had to pay attention
to a determined stimulus, the ear was stimulated for ten seconds
only in the side where attention should be focused on. This was
performed in order to instruct the subject on the cognitive task.
Fig. 1 illustrates the stimulation protocol.

vector and the horizontal median line crossing the head, as illustrated in Fig. 2.
Each vector is then weighed by the respective amount of MSC in
f0 , where f0 is either 38 Hz (right stimulus) or 32 Hz (left stimulus).
This procedure results in a set of vectors that are summed together
to result in a single vector called spatial coherence, as follows:

2.2. Magnitude-squared coherence (MSC)

 spatial (f0 ) =

The magnitude-squared coherence (MSC) estimate between

two random, discrete-time signals, x[n] and y[n], can be estimated with the familiar approach of partitioning them into M
non-overlapping segments as


2
 M

 i=1 Xi (f )Yi (f )
2
 xy (f ) = M
M
2
i=1

|Xi (f )|

i=1

|Yi (f )|2

(1)

where Xi (f) and Yi (f) are, respectively, the ith-window Fourier

Transform of x[n] and y[n] and and * superscript mean, respectively, estimation and complex conjugate.
The MSC between the stimulation signal and the EEG has been
proposed by Dobie and Wilson (1989) as a way of testing for the
presence of evoked responses. The MSC, which is derived from the
coherence function in Eq. (1) between a periodic, deterministic signal x[n] and a random one y[n] can be estimated using only the latter
as:


2
 M

 i=1 Yi (f )
2
MSC =  (f ) = M
.
2
M

i=1

|Yi (f )|

(2)

2.3. Spatial coherence

MSC is an objective detection technique that indicates the
degree of linear dependence between the harmonic components
of the stimulus and the response obtained in the EEG (Dobie and
Wilson, 1989). In the case of an ER during an ASA task, it is also
interesting to detect the attention focus; i.e. the right or left ear.
In other words, it is necessary to determine the scalp regions with
stronger synchronism with the stimuli and to establish to which
ear such locations are related. With this aim, spatial coherence is
then dened as a spectral measure that takes into account both the
MSC-value in each electrode and its spatial location on the scalp.
Initially, in order to dene the approximate electrodes location,
the head is considered as unit circle. The position of the electrodes
in such circle is dened in accordance with 1020 International
System and was obtained by the toolbox EEGLAB (Delorme and Anf
Makeig, 2004). Each electrode is then considered as a vector, where
module Ai with i ranging from 1 to the number of EEG derivations
used in the estimation - is the distance from this electrode to the
vertex (Cz ) and the phase ( i ) is the angle formed between each

N


i Ai eji

(3)

i=1

where N represents the number of channels used for the calculation of spatial coherence, Ai is the module and  i is the angle of
each electrode, as shown in Fig. 2. The amount of MSC  is equal
to  2 (f0 )attention  2 (f0 )no-attention and the subscripts no-attention
and attention are the MSC estimated before and after the selective
attention protocol (refer to Fig. 4 for further details), respectively.
Thus, spatial coherence is a resulting vector that will point out to
the region of the scalp with greater synchronism for the frequency
f0 in analysis.
In order to obtain a binary classication of the focus of attention,
i.e. to the left or to the right, the attention index was developed
based on the spatial coherence and is shown in Eq. (4). In fact, the
attention index is calculated by taking only the real part, i.e. projection on the x axis. Thus, it only considers inter-hemispherical
changes and was developed based on reports that attention causes
a contralateral modulation in the ASSR (Ross et al., 2004; Gander
et al., 2007; Muller et al., 2009). Hence, if the index results in a positive value (spatial pointing to the right), it indicates attention in
the left stimulus (contralateral). Conversely, if the index is negative
(pointing to the left), it indicates attention in the right stimulus.
index = Real( spatial )

(4)

Fig. 3 exemplies the application of the attention index calculation. The dotted arrows indicate each vector formed by the amount

Fig. 2. Scalp vectors formed by each electrode and vertex, taking module (Ai ) and
angle (i ) to calculate spatial coherence, where i last from 1 to the number of EEG
derivations.

L. Bonato Felix et al. / Journal of Neuroscience Methods 229 (2014) 2832

31

Spatial coherence is a method that considers both the position

of the electrodes and the MSC values. Hence, it is clearly affected by
the set of electrodes used. Thus, in this study, all possible electrode
combinations were evaluated by exhaustive search throughout the
subjects. The combination that led to the best mean hit rate was
found to be the channels T3, F3, P3, O1, T4, P4, O2, Fz and Pz.
3. Results and discussion
3.1. Spatial contralateralization of ASSR in attention condition

Fig. 3. Example of application of attention index calculation.

of MSC and electrode position; spatial coherence is the resultant

vector (solid arrow). Fig. 3 also shows the attention index, which
in the example has a positive value, indicating attention to the left
stimulus.
2.4. Evaluation of spatial coherence for BCI applications
In order to evaluate the use of spatial coherence to detect the
side of the auditory spatial attention, two metrics have been used,
namely the number of commands that might be given per minute
(speed) and the hit hate. The rst one was obtained considering different fractions of the stimulation period shown in Fig. 1. Thus, the
use of the whole stimulation period without pause (2 min), would
result in a 1-command per minute speed. The hit hate was obtained
as the success rate for such speed.

Ross et al. (2004), Gander et al. (2007), and Muller et al. (2009)
reported evidences of the spatial modulation of ASSR with attention
status. For a spatial evaluation of these responses, topographical
maps of the scalp were represented in Fig. 4. This gure shows the
condition without attention (top maps) and the difference between
the attention condition and without attention focus (bottom maps),
obtained from the MSC mean value, for all individuals, calculated
for each electrode, in each stimulation frequency.
In the topographical maps of Fig. 4, for the no attention condition, a well-dened region of higher synchronism is not noticed for
left and right modulation frequencies. However, in the attention
condition (bottom panel of Fig. 4), considering the stimulating frequency to the left ear (32 Hz), a region of higher coherence values is
observed in the right hemisphere. The same occurs for the right ear
frequency (38 Hz), where a region with greater coherence values
is observed in the left hemisphere during attention, mainly in the
fronto-central region. These results indicate a contralateral modulation of ASSR in the attention condition and are in agreement with
prior studies.

Fig. 4. Mean scalp topographies in conditions without attention and attention on stimuli. These scalp maps show the distribution of MSC-values in the left and right
steady-state frequencies.

32

L. Bonato Felix et al. / Journal of Neuroscience Methods 229 (2014) 2832

Table 1
Hit rate for number of commands per minute.

References

Commands per minute

Hit rate (%)

23.50
14.10
10.07
7.83
6.41
5.87
5.03
4.41
4.15
3.92
3.36

50
54
64
71
71
79
82
82
79
75
75

3.2. Hit rate and speed

Table 1 shows the hit rate and commands per minute obtained
with the spatial coherence as described in Section 2.4. Since spatial
coherence is obtained as the sum of vectors with amplitude given by
the MSC-values, it is strongly affected by the number M of segments
used in expression (2). Hence, the speed and hit rate also change
with M. Using segments with 1024 points each (i.e. with 1.707 s
for the 600 Hz sampling frequency), the best case with speed = 5.03
commands per minute and hit hate = 82% occurred for M = 89. Using
a similar paradigm, Kim et al. (2011) obtained an equivalent performance (maximum hit rate of 84.33%, for a classication time of
20 s 3 commands per minute).
4. Conclusions
The present work proposed a new method for detecting the
focus of selective attention, spatial coherence, and it seemed useful as a classier of the attention condition. The obtained hit rates
may be considered appropriate, with a maximum value of 82 %
(comparable with the similar method of Kim et al. (2011)). The
method considers spatial distribution of electrodes, therefore other
arrangements, even unconventional ones and with higher location
accuracy, can improve the performance of the method.
The results also provide evidences on the contralateral modulation of ASSR, in the attention condition. Thus, for binaural
stimulation, with AM tones of different frequencies, the responses
to the focused stimulus have exhibited increased coherence values
in the contralateral hemisphere of the ear to which the individual paid attention. These results are in agreement with previous
studies developed by Ross et al. (2004), Gander et al. (2007) and
Muller et al. (2009), which propose a spatial modulation of ASSRs
in attention conditions.
Acknowledgements
The authors acknowledge FAPEMIG, FAPERJ, CAPES and CNPq
for nancial support.

Bronkhorst AW. The cocktail party phenomenon: A review of research on speech

intelligibility in multiple-talker conditions. Acustica 2000;86:11728.
Chiappa KH. Evoked potentials in clinical medicine. New York: Raven Press;
1997.
Cohen LT, Rickards FW, Clark GM. A comparison of steady-state evoked potentials to modulated tones in awake and sleeping humans. J Acoust Soc Am
1991;90(5):246779.
Delorme A, Anf Makeig S. EEGLAB: an open source toolbox for analysis of single-trial
EEG dynamics. J Neurosci Methods 2004;134:921.
Dobie RA, Wilson MJ. Analysis of auditory evoked responses by magnitude squared
coherence. Ear Hear 1989;10:213.
Dobie RA, Wilson MJ. Objective response detection in the frequency domain. Electroencephalogr Clin Neurophysiol 1993;88:51624.
Felix LB, Moraes JE, Miranda de S AMFL, Yehia HC, Moraes MFD. Avoiding spectral
leakage in objective detection of auditory steady-state evoked responses in the
inferior colliculus of rat using coherence. J Neurosci Methods 2005;144(June
(2)):24955.
Gander P, Bosnyak D, Wolek R, Roberts LE. Modulation of the 40 Hz auditory
steady-state response by attention during acoustic training. Int Cong Ser
2007;1300:3740.
Giard M, Fort A, Mouchetant-Rostaing Y, Perrier J. Neurophysiological mechanisms of auditory selective attention in humans. Front Biosci 2000;5:
8494.
Joris PX, Schreiner CE, Rees A. Neural processing of amplitude-modulated sounds.
Physiol Rev 2004;84:54177.
Kim DW, Hwang HJ, Lim JH, Lee YH, Jung KY, Im CH, et al. Classication of selective
attention to auditory stimuli: Toward vision-free brain computer interfacing. J
Neurosci Methods 2011;197(1):1805.
Kuwada S, Batra R, Maher VI. Scalp potentials of normal and hearing-impaired
subjects in response to sinusoidal amplitude modulated tones. Hear Res
1986;21(2):17992.
Levi EC, Folsom RC, Dobie RA. Amplitude modulation following response (AMFR)
effects of modulation rate, carrier frequency, age and state. Hear Res
1993;68(1):4252.
Linden RD, Picton TW, Hamel G. Human Auditory Steady-State Evoked Potentials
during Selective Attention. Electroen- cephalography and Clinical Neurophysiology 1987;2(66):14559.
Melges DB, Infantosi AFC, Miranda De S AMFL. Topographic distribution of the
tibial somatosensory evoked potential using coherence. Braz J Med Biol Res
2008;41(December (12)):105966.
Miranda De S AMFL, Felix LB. Improving the detection of evoked responses to
periodic stimulation by using multiple coherence application during photic
stimulation. Med Eng Phys 2002;24(March (4)):24552.
Muller N, Schlee W, Hartmann T, Lorenz I, Weisz N. Top-down modulation of the
auditory steady-state response in a task-switch paradigm. Front Hum Neurosci
2009;3:19.
Perry NW. Signal versus Noise in the Evoked Potential. Science 1966;153:
1922.
Plourde G, Picton TW. Human auditory steady-state response during general anesthesia. Anesth Anal 1990;71(November (5)):4608.
Rickards FW, Tan LE, Cohen LT, Wilson OJ, Drew JH, Clark GM, et al. Auditory
steady-state evoked potential in newborns. Br J Audiol 1994;28(December (6)):
32737.
Ross B, Draganova R, Picton TW, Pantev C. Frequency specicity of 40-Hz auditory
steady-state responses. Hear Res 2003;186(1):5768.
Ross B, Picton TW, Herdman AT, Pantev C. The effect of attention on the auditory
steady-state response. Neurol Clin Neurophysiol 2004(22):14.
Van Dun B, Wouters J. Optimal electrode selection for multi-channel electroencephalogram based detection of auditory steady-state responses. J Acoust Soc
Am 2009;126:25468.
Wolpaw JR, Birbaumer N, Mcfarland DJ, Pfurtscheller G, Vaughan TM. Brain
Computer Interfaces for communication and control. Clin Neurophysiol
2002;113:76791.