Neuroscience 250 (2013) 49–59

THE EFFECTS OF MUSIC ON BRAIN FUNCTIONAL NETWORKS: A

NETWORK ANALYSIS
J. WU, a,b J. ZHANG, a,b X. DING, c R. LI a,b AND using graph theoretical measures (Stam and Reijneveld,
C. ZHOU a,b* 2007; Bullmore and Sporns, 2009; He and Evans, 2010;
a
Cognitive Science Department, Xiamen University, Xiamen, China Bullmore and Bassett, 2011). One effective model is
b
Fujian Key Laboratory of the Brain-like Intelligent Systems, small-world networks in terms of graph theory. Small-
Xiamen University, Xiamen, China world networks appear as regular, lattice-like structures
c
College of Foreign Languages and Cultures, Xiamen
with a few random long-range connections (Watts and
University, Xiamen, China Strogatz, 1998). They are characterized by high local
clustering, signifying dense local connectivity, and short
path length between any two vertices. Small-world
Abstract—The human brain can dynamically adapt to the organizations are interesting for complex brain networks
changing surroundings. To explore this issue, we adopted since they confer high local and global efficiency of
graph theoretical tools to examine changes in electroen- information transfer with a low wiring cost (Bullmore and
cephalography (EEG) functional networks while listening
Sporns, 2009; He and Evans, 2010).
to music. Three different excerpts of Chinese Guqin music
There is increasing evidence for small-world
were played to 16 non-musician subjects. For the main fre-
quency intervals, synchronizations between all pair-wise configurations in structural networks of animals (Sporns
combinations of EEG electrodes were evaluated with phase and Zwi, 2004; Sporns et al., 2007). These
lag index (PLI). Then, weighted connectivity networks were configurations are extended to studies of human brain
created and their organizations were characterized in terms structural networks based on cortical thickness
of an average clustering coefficient and characteristic path measurements (He et al., 2007), diffusion tensor imaging
length. We found an enhanced synchronization level in the (DTI) (Iturria-Medina et al., 2008; Yan et al., 2011) and
alpha2 band during music listening. Music perception diffusion spectrum imaging (DSI) (Hagmann et al.,
showed a decrease of both normalized clustering coefficient 2008). In several functional magnetic resonance imaging
and path length in the alpha2 band. Moreover, differences in
(fMRI) studies, networks of functional connectivity also
network measures were not observed between musical
exhibited small-world characteristics (He et al., 2009;
excerpts. These experimental results demonstrate an
increase of functional connectivity as well as a more random Hayasaka and Laurienti, 2010; Tian et al., 2011). While
network structure in the alpha2 band during music percep- neuronal activities frequently occur in time-scales shorter
tion. The present study offers support for the effects of than those reflected in fMRI signal fluctuations (Palva
music on human brain functional networks with a trend et al., 2010), the use of electroencephalography (EEG)
toward a more efficient but less economical architecture. and magnetoencephalography (MEG) guarantees an
Ó 2013 IBRO. Published by Elsevier Ltd. All rights reserved. adequate temporal resolution. And such dynamic
techniques are inherently appropriate to detect
oscillatory synchronization (Sporns et al., 2000), which
Key words: electroencephalography, music, graph theory,
takes a crucial role in interregional communication
small-world network, synchronization.
(Fries, 2005; Schnitzler and Gross, 2005). Network
analyses of EEG and MEG are consistent with the fMRI
data, indicating small-world features in synchronization
INTRODUCTION patterns (Stam et al., 2007a, 2009; Deuker et al., 2009;
The brain is considered to be a complex system, Douw et al., 2011; Jin et al., 2012). The small-world
comprised of spatially interconnected areas. Structural network attributes could be found not only in resting but
and functional connectivity of the brain possess also during task conditions, such as finger tapping and
properties of complex networks and can be investigated music listening (Eguı́luz et al., 2005). Interestingly, brain
maturation may be associated with a shift toward a more
*Correspondence to: C. Zhou, Cognitive Science Department, Xia- small-world network architecture (Boersma et al., 2011).
men University, 422 South Siming Road, Xiamen 361005, China. Tel: In contrast, brain networks deviate from the optimal,
+86-13850083875; fax: +86-592-2580168. normal small-world organization in brain pathology (Stam
E-mail address: [email protected] (C. Zhou). et al., 2007a, 2009; Rubinov et al., 2009).
Abbreviations: DSI, diffusion spectrum imaging; DTI, diffusion tensor
imaging; EEG, electroencephalography; EOG, electrooculogram; FIR,
It has been suggested that brain networks exist in a
finite impulse response; fMRI, functional magnetic resonance imaging; critical state between chaos and order (Kitzbichler et al.,
GLS, GuangLingSan; MEG, magnetoencephalography; MHSN, 2009; Petermann et al., 2009). Small-world properties
MeiHuaSanNong; OLWJ, OuLuWangJi ; PLI, phase lag index; PSD, support rapid adaptive reconfiguration of functional
power spectral density; SD, standard deviation.

0306-4522/13 $36.00 Ó 2013 IBRO. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.neuroscience.2013.06.021
49
50 J. Wu et al. / Neuroscience 250 (2013) 49–59

connectivity in response to varying cognitive demands difference between two time series. It can determine a
(Bassett et al., 2006). Several empirical studies have reliable evaluation of synchronization that is less
identified task-related changes in network organization sensitive to the influence of volume conduction.
during finger movements (Bassett et al., 2006; Jin et al., Therefore we utilize the PLI to quantify functional
2012), working memory (Palva et al., 2010; Kitzbichler connectivity.
et al., 2011) and learning (Bassett et al., 2011). In this study, we intend to further explore the effects of
Another potential task that can cause network music on EEG functional networks. Our questions are: (i)
reconfiguration is music perception. Music is widely whether network reorganization takes place during music
recognized as a universal language. Music in various perception in different frequency bands and (ii) whether
forms has a profound impact on all cultures, and even differences in network organization can be observed
on animals (Gray et al., 2001). Listening to music is a between musical pieces. The experiment used three
complex cognitive task and implicates processing and pieces of Chinese Guqin music as musical stimuli.
integration of meaningful elements including harmony, Guqin music is regarded as a symbol of Chinese
rhythm or melody (Schmithorst, 2005). Modulating civilization (Zhu et al., 2008, 2009) and has been
effects of music have been observed on functional selected as a Masterpiece of the Oral and Intangible
connectivity such as default mode network (DMN) (Kay Heritage of Humanity by UNESCO. Cognitive
et al., 2012) and oscillatory synchronization (Petsche neuroscientific research on Guqin music perception will
et al., 1997; Bhattacharya et al., 2001; Bhattacharya deepen our understanding of the perception of Eastern
and Petsche, 2005; Flores-Gutiérrez et al., 2007, 2009; music.
Ruiz et al., 2009). Wu et al. (2012) have identified a
topological change in EEG alpha-band networks in a EXPERIMENTAL PROCEDURES
music listening task while preserving the small-world
characteristics comparable to the findings of Eguı́luz Participants and materials
et al. (2005). This study involved 16 right-handed subjects (mean age
While the previous study has reported network 22.25 years, standard deviation (SD) 1.65; eight males).
reconfiguration during music perception, three issues All subjects were recruited from the university campus
demand further attention: (i) the topological differences and nurtured in China. Structured interviews guaranteed
between musical excerpts have not been considered; (ii) that the subjects had normal hearing, no formal or
the earlier analysis was focused merely on the alpha informal training in music, and were free from
band, whereas the functional networks were usually neurological diseases or psychoactive drugs use. Each
probed in multiple frequency intervals (Deuker et al., subject was informed of the experimental procedure and
2009; Stam et al., 2009; Boersma et al., 2011; gave a written consent to participate. The subjects were
Kitzbichler et al., 2011; Jin et al., 2012) since different paid for the participation in the experiment.
frequency components might represent different All subjects were instructed to listen attentively to
biological significances; (iii) estimation of brain functional three pieces of Guqin music: OuLuWangJi (OLWJ),
connectivity from EEG recordings can be affected by MeiHuaSanNong (MHSN), and GuangLingSan (GLS)
volume conduction, which may make interpretation (music conditions). These selected musical pieces were
unreliable. unfamiliar to all participants. In addition, subjects also
The problem of volume conduction refers to the listened to a segment of pink noise (noise condition).
inclination of nearby EEG electrodes to pick up activity The presentation of each acoustic stimulus lasted 40 s
of common neural sources, introducing high synchronies and their order was counterbalanced among subjects.
that do not reflect factual functional connectivity (Nunez EEG at rest was recorded for 2 min and used as a
et al., 1997; Stam et al., 2007b). There are two main control condition (silence condition). The subjects were
approaches proposed to tackle this problem. A first comfortably seated inside a dimly illuminated and sound
approach attempts to use reconstructed sources as a attenuated chamber. The sound backgrounds were
basis to assess functional correlations (Gross et al., presented in two stereo speakers located at 2 m in front
2001; Lehmann et al., 2006; Amor et al., 2009). of the subjects. The volume arriving at the subjects was
Although this method indeed makes a contribution to adjusted to 60 dB SPL. Subjects were asked to keep
detecting functional co-operations between anatomically their eyes closed during the experiment.
well-specified brain areas, a primary defect is the
absence of a unique solution to the EEG inverse
EEG recordings
problem. A second approach is to determine true
interactions between time series which are not likely to EEG was recorded using the Neuroscan system
be biased by volume conduction. One could adopt the (Neuroscan Inc., El Paso, TX) with 64 electrodes
imaginary part of the complex coherency, which is not arranged according to the international 10–20 system.
owing to volume conduction effects (Nolte et al., 2004). All leads were referenced to left and right earlobes. To
However, it contains mixed information on both monitor eye movements and blinks, horizontal and
amplitudes and phase delays, and consequently is not vertical electrooculograms (EOGs) were recorded from
an appropriate measure of functional correlations. Stam two bipolar electrodes placed slightly lateral to the outer
et al. (2007b) proposed phase lag index (PLI) as an canthus of each eye and above and below the left eye.
alternative. The PLI estimates consistency of phase Inter-electrode impedance levels were kept below 5 kX.
 J. Wu et al. / Neuroscience 250 (2013) 49–59 51

EEG was continuously recorded with a 0.05–100 Hz where function signðuab ðnÞÞ returns 1 if uab ðnÞ is positive, and
bandpass filter, at a 1000 Hz sampling frequency. returns 1 if uab ðnÞ is negative; N denotes the amount of
Ocular artifacts were reduced by means of a regression sampling points. The value of PLIab is restricted between 0 and
approach (Semlitsch et al., 1986) implemented in Scan 1. It reaches 0 for no co-ordination or coupling with a relative
4.3 (Neuroscan Inc., El Paso, TX). In order to avoid phase that encircled 0 mod p, which is likely to result from
transition effects, the first and last 5-s data of each volume conduction. Whereas PLIab = 1 in the condition of strict
phase locking at a constant, nonzero lag.
music and noise segments were discarded from the
stored EEG. Thus, for each subject, we chose one The PLI values were averaged over all pair-wise
artifact-free epoch of 30 s (30,000 samples) for each combinations of EEG electrodes to indicate the mean
task as well as for rest. Using a causal finite impulse levels of phase synchronization of cortical areas. In
response (FIR) filter (24 dB/octave roll-off), we digitally addition to computing the global mean of PLI, a more
filtered the signal in each epoch into the classic detailed, regional approach was adopted. For this, EEG
frequency bands: delta (0.1–4 Hz), theta (4–8 Hz), electrodes were grouped into frontal, central, parietal,
alpha1 (8–10 Hz), alpha2 (10–13 Hz), beta (13–30 Hz), occipital and temporal areas for both hemispheres.
and gamma (30–45 Hz). Averaging was done to obtain intraregional (short-
distance) and interregional (long-distance) PLI.
Power spectral analysis Intraregional PLI was calculated as the mean value of PLI
for all electrode pairs within one area, and interregional
For each epoch and every electrode, the power spectral PLI was calculated between electrodes from two different
density (PSD) of EEG was obtained via Welch’s method areas. Midline channels were not involved.
of spectral estimation (eight equal-length sections with
50% overlap, Hamming window). The PSD values were
averaged across all EEG channels, resulting in a single Network analysis
value for each band. Power spectral analysis, A network can be typically represented by a graph,
calculations of the PLI, network analysis and statistical defined as a set of vertices linked by edges. In this
analysis were implemented in Matlab (MathWorks Inc., study, we constructed weighted graphs with 62 vertices
Natick, MA). (corresponding to the 62 available EEG channels) and
employed the matrices of PLI values for all pairs of
Calculation of phase lag index EEG channels as edge weights. That is, we used
We quantified synchronizations between all pair-wise 62  62 correlation matrices, and the value of each
combinations of EEG channels via the PLI. The PLI element wij was assigned as the weight of the edge
extracts consistency of phase synchrony between two between two vertices i and j. The procedure involved in
signals. It is less sensitive to the contribution from weighted graph analysis applied to the EEG data is
volume conduction and thus can provide a reliable shown in Fig. 1.
appraisal of coupling strength (Stam et al., 2007b; The clustering coefficient CW and the characteristic
Pearce et al., 2010). Moreover, the PLI is inherently path length LW were utilized to quantify the topological
suitable to deal with nonlinear and non-stationary properties of weighted graphs. We adopted weighted
signals like EEG. network measures so as to make full use of information
For any arbitrary EEG signal x(t), the analytic signal in the weights and avoid the arbitrary selection of
w(t) is constructed by a complex function of time: threshold for PLI values. See Stam and Reijneveld
Z þ1 (2007) for an in-depth discussion.
xðsÞ
wðtÞ ¼ xðtÞ þ ixH ðtÞ  xðtÞ þ ip1 p:v: ds; ð1Þ In principle, the clustering coefficient for a vertex i
1 ts
quantifies the proportion of its neighboring vertices j that
where xH ðtÞ is the Hilbert transform of x(t), and p.v. denotes the are also connected to each other. Its calculation from
Cauchy principal value. Hilbert transform is virtually the weighted graphs demands weight symmetry (wij = wji)
convolution of x(t) with 1/pt. The instantaneous amplitude A(t) and 0 6 wij 6 1. These two conditions are satisfied when
and instantaneous phase /ðtÞ of x(t) can be derived in terms of employing PLI as edge weights. The clustering
the analytic signal’s polar form: coefficient CW of vertex i can be formalized as (Stam
i
wðtÞ ¼ AðtÞei/ðtÞ : ð2Þ et al., 2009)
P P
In turn, the phase is uniquely determined as j–i k–i wij wik wjk
CW k–j
¼ P P : ð6Þ
xH ðtÞ i
k–i wij wik
/ðtÞ ¼ arctan : ð3Þ j–i k–j
xðtÞ
CW is the mean clustering coefficient over all vertices,
From the phases of two EEG signals xa(t) and xb(t), computed as
the phase difference or relative phase is formed as
uab ðtÞ ¼ /a ðtÞ  /b ðtÞ: ð4Þ 1X M
CW ¼ CW ; ð7Þ
M i¼1 i
Then, the PLI is defined as an asymmetry measure for
the phase difference distribution via where M denotes the number of vertices. CW depicts the local
 
1 X
N1  density of the entire graph.
 
PLIab ¼ signðuab ðnÞÞ; ð5Þ We applied the notion of the global efficiency E
N n¼0 
(Latora and Marchiori, 2001) to compute the path length
52 J. Wu et al. / Neuroscience 250 (2013) 49–59

Procedure for weighted graph analysis

A B

calculation of
phase lag index

D C FP1-ch.1
0.6 0.6
F2-ch.11
0.5 0.5
FC4-ch.21
0.4 0.4
C6-ch.31
0.3 0.3
TP8-ch.41
0.2 0.2
PO7-ch.51
0.1 0.1

O2-ch.61 0
0
ch.1 ch.11 ch.21 ch.31 ch.41 ch.51 ch.61

randomization
CW LW

E 0.6

0.5
F γ=CW/CW-s C -s
W
0.4

λ=LW/LW-s 0.3
LW-s
0.2

0.1

Fig. 1. Schematic representation of the procedure involved in weighted graph analysis of EEG recordings. EEG signals are recorded from
electrodes illustrated in (A). Epochs of EEG data are filtered within frequency bands of interest (B). PLI is estimated as a measure of correlation
between all pairs of channels, resulting in correlation matrices (C). Then, weighted graphs are constructed based on the correlation matrices (D).
The clustering coefficient CW and characteristic path length LW are calculated to characterize each graph. In addition, surrogate graphs are derived
by randomly shuffling the cells of the correlation matrices (E). For each original graph, network parameters are assessed and averaged over the
ensemble of surrogate graphs, resulting in CW-s and LW-s. Finally, the normalized parameters CW/CW-s and LW/LW-s are determined (F).

of weighted graphs. The length of an edge is determined Each network has different basic characteristics, such
as the reciprocal of its edge weight, i.e., Lij = 1/wij if as structure, size, and edge weights, influencing both
wij – 0, and Lij = 1 if wij = 0. The distance dij between values of CW and LW. In order to derive network
the vertices i and j is the minimum of the sum of edge parameters that are independent of differences in the
lengths between these two vertices. The characteristic PLI, CW and LW were compared to the average values
path length LW of the whole graph is the average dij for for an ensemble of surrogate graphs. Fifty surrogate
all possible pairs of vertices. This definition employs the graphs were constructed from each original graph by
harmonic mean to handle the problem of disconnected randomly shuffling the edge weights. The normalized
graphs (Newman, 2003), for which dij ? 1. That is, parameters were calculated as
1 1 c ¼ CW =CW -s and k ¼ LW =LW -s; ð9Þ
LW ¼ ¼ P PM 1 : ð8Þ
E ð1=MðM  1ÞÞ Mi¼1 j¼1 d
ij
j–i
where CW-s and LW-s denoted the mean CW and LW over 50
W
L reflects the overall integration of the graph. surrogate random graphs.
 J. Wu et al. / Neuroscience 250 (2013) 49–59 53

Table 1. Average PSD while exposure to five sound backgrounds in six frequency bands

Bands OLWJ MHSN GLS Noise Silence

Mean ± SD Mean ± SD Mean ± SD Mean ± SD Mean ± SD

Delta 40.6767 ± 24.4940 46.0428 ± 24.0124 39.8405 ± 35.8843 60.9527 ± 43.5244 64.5141 ± 43.1975
Theta 1.4757 ± 0.8573 1.3639 ± 0.9402 1.4023 ± 0.9736 1.6919 ± 0.9183 2.3405 ± 1.8314
Alpha1 2.5426 ± 2.0330 2.4717 ± 1.9757 2.5167 ± 2.0821 2.2861 ± 1.8435 2.5738 ± 1.7728
Alpha2 1.1471 ± 0.4437 1.0981 ± 0.5016 0.9555 ± 0.3833 0.9754 ± 0.3794 1.0923 ± 0.4162
Beta 0.2640 ± 0.0740 0.2582 ± 0.0790 0.2598 ± 0.0777 0.3473 ± 0.1296 0.4207 ± 0.2848
Gamma 0.0253 ± 0.0106 0.0233 ± 0.0081 0.0247 ± 0.0077 0.0393 ± 0.0127 0.0422 ± 0.0313

Weighted EEG networks

OLWJ MHSN GLS Noise Silence
0.3
0.25
0.2
Delta

0.15
0.1
0.05
0
0.25
0.2
Theta

0.15
0.1
0.05
0
0.4
0.35
0.3
Alpha1

0.25
0.2
0.15
0.1
0.05
0
0.4
0.35
0.3
Alpha2

0.25
0.2
0.15
0.1
0.05
0
0.25
0.2
Beta

0.15
0.1
0.05
0
0.1
0.09
0.08
Gamma

0.07
0.06
0.05
0.04
0.03
0.02
0.01
0

Fig. 2. Mean weighted networks in six frequency bands for five sound backgrounds: OLWJ, MHSN, GLS, noise, and silence. The strength of PLI for
all pairs of EEG channels is indicated with a jet scale, from blue (PLI = 0) to red (high PLI). (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)

Statistical analysis comparisons was applied (i.e., P values < 0.05/

10 = 0.005 were considered as significant).
Since parametric statistical analyses might not be
applicable for the subject size, non-parametric methods
were carried out. The PSD, PLI and network measures
among background conditions were compared by
RESULTS
Wilcoxon signed rank test. As the possible 10 Statistical comparisons for average PSD during exposure
correlations among five sound backgrounds to five sound backgrounds were performed with the non-
(C25 ¼ 5  4=2 ¼ 10) were subjected to multiple non- parametric Wilcoxon signed rank test in six different
independent tests, Bonferroni correction for multiple frequency bands. The results are shown in Table 1. No
54 J. Wu et al. / Neuroscience 250 (2013) 49–59

significant differences were detected among the sound and GuangLingSan (P = 0.0032) excerpts in relation to
backgrounds. the noise condition. Increases within alpha2 frequency
Fig. 2 shows the mean graphs for different were found during listening to OuLuWangJi
background conditions calculated with phase lag index (P = 0.0045), MeiHuaSanNong (P = 0.0045) and
in six frequency bands. Sketchy differences between the GuangLingSan (P = 0.0023) as compared to the
music and noise conditions and between the music and silence condition. Although OLWJ showed a higher
silence conditions could be observed via visual scrutiny, value in comparison with the noise condition
particularly in the alpha1 and alpha2 bands. (P = 0.0052) in the alpha2 band, this difference did not
The mean PLI values averaged across all pairs of reach significance after Bonferroni correction for multiple
EEG channels are shown in Table 2. There were comparisons (a = 0.05/10 = 0.005). In addition, similar
significant differences mainly in the alpha2 band. The non-significant trends were observed in the alpha1
synchronization levels for alpha2 band were significantly band. The mean PLI values were higher in the music
higher during listening to MeiHuaSanNong (P = 0.0019) conditions than in the noise condition: for OLWJ

Table 2. Mean PLI during exposure to five sound backgrounds in six frequency bands

Bands OLWJ MHSN GLS Noise Silence

Mean ± SD Mean ± SD Mean ± SD Mean ± SD Mean ± SD

Delta 0.1281 ± 0.0188 0.1340 ± 0.0162 0.1386 ± 0.0269 0.1520 ± 0.0260 0.1548 ± 0.0196
Theta 0.0983 ± 0.0334 0.1267 ± 0.0486 0.1149 ± 0.0454 0.0954 ± 0.0265 0.0882 ± 0.0241
Alpha1 0.2245 ± 0.0738 0.2380 ± 0.0862 0.2241 ± 0.0801 0.1632 ± 0.0485 0.1497 ± 0.0434
Alpha2 0.2328b ± 0.0702 0.2452a,b ± 0.0834 0.2300a,b ± 0.0779 0.1566a ± 0.0483 0.1499b ± 0.0406
Beta 0.0986 ± 0.0322 0.1118 ± 0.0401 0.1078 ± 0.0364 0.0767 ± 0.0233 0.0733 ± 0.0206
Gamma 0.0354 ± 0.0048 0.0363 ± 0.0054 0.0340 ± 0.0046 0.0338 ± 0.0042 0.0331 ± 0.0051
For each band, letter superscripts indicate statistically significant differences between backgrounds (Wilcoxon signed rank test, P < 0.005, Bonferroni-corrected).
a
Significant differences between music and noise conditions.
b
Significant differences between music and silence conditions.

Regional differences in phase lag index

Alpha2
OLWJ vs. Noise MHSN vs. Noise GLS vs. Noise

F F F F F F

T C C T T C C T T C C T

P P P P P P

O O O O O O

OLWJ vs. Silence MHSN vs. Silence GLS vs. Silence

F F F F F F

T C C T T C C T T C C T

P P P P P P

O O O O O O

Fig. 3. Robust differences in interregional (denoted by solid lines) and intraregional (denoted by filled squares) PLI for the alpha2 band (Wilcoxon
signed rank test, P < 0.005). The statistical results can only be considered as indicative findings. Solid lines or filled squares in red: sound
backgrounds in red higher than backgrounds in blue; solid lines or filled squares in blue: backgrounds in blue higher than those in red. In comparison
with the noise and silence conditions, the music conditions displayed increased fronto-parietal, fronto-temporal, centro-parietal, centro-occipital and
temporo-parieto-occipital PLI. Intraregional synchronizations for frontal, central and temporal areas were also higher in the music conditions. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
 J. Wu et al. / Neuroscience 250 (2013) 49–59 55

Network differences (alpha2 band)

P8
OLWJ vs. Noise MHSN vs. Noise GLS vs. Noise OLWJ vs. MHSN OLWJ vs. GLS

OLWJ vs. Silence MHSN vs. Silence GLS vs. Silence MHSN vs. GLS Noise vs. Silence

P14
OLWJ vs. Noise MHSN vs. Noise GLS vs. Noise OLWJ vs. MHSN OLWJ vs. GLS

OLWJ vs. Silence MHSN vs. Silence GLS vs. Silence MHSN vs. GLS Noise vs. Silence

Fig. 4. Topographical patterns that exhibit considerable differences in PLI between sound backgrounds within the alpha2 band. The single subjects
P8 and P14 are shown. Solid lines appear only when the absolute values of the corresponding differences are greater than a threshold T. T (= 50%
of the maximum for PLI) was selected to make enough connections visible. For P8, T = 0.3508; for P14, T = 0.3383. Solid lines in red: sound
backgrounds in red higher than backgrounds in blue; solid lines in blue: backgrounds in blue higher than those in red. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

(P = 0.0097), MHSN (P = 0.0131) and GLS Fig. 4 presents topographical patterns at participant
(P = 0.0151). The values of PLI showed increases for level, which affords further insight into the network
OLWJ (P = 0.0097), MHSN (P = 0.0131) and GLS differences within the alpha2 band. In general, the
(P = 0.0052) in relation to the silence condition. differences between the music and noise conditions and
Fig. 3 shows regional results for the alpha2 frequency, between the music and silence conditions were in the
in which statistical differences in global mean PLI could be same direction (i.e., more increases for music conditions
found, to provide further illustration. Instead of multiple when compared with the noise and silence conditions),
comparisons adjustment, a predetermined fixed although the individual patterns varied. For the
threshold was selected to act as a statistical filter (i.e., comparisons between the music conditions and
P < 0.005). Hence the statistical results as shown in between the noise and silence conditions, the
Fig. 3 can merely be regarded as indicative findings. In differences moved in all possible directions, and
relation to the noise and silence conditions, the music therefore averaged out when combined.
conditions showed an increase in fronto-parietal, fronto- The mean values of the clustering coefficient CW and
temporal, centro-parietal, centro-occipital and temporo- the characteristic path length LW are presented in Table
parieto-occipital PLI. Local frontal, central and temporal 3. There were significant differences in the alpha2
synchronizations were also higher in the music frequency band. The Wilcoxon signed rank test
conditions. No obvious distinctions could be found demonstrated that CW was higher during listening to
between the music conditions and between the noise OLWJ (P = 0.0023), MHSN (P = 0.0038) and GLS
and silence conditions. (P = 0.0013) in comparison with the noise condition. CW
56 J. Wu et al. / Neuroscience 250 (2013) 49–59

showed higher values for OLWJ (P = 0.0032), MHSN listening to OLWJ (P = 0.0032), MHSN (P = 0.0023)
(P = 0.0045) and GLS (P = 0.0038) as compared with and GLS (P = 0.0045) in comparison with the noise
the silence condition. For LW decreases were found condition. k was found to be shorter for OLWJ
during listening to OLWJ (P = 0.0007), MHSN (P = 0.0038), MHSN (P = 0.0027) and GLS
(P = 0.0013) and GLS (P = 0.0019) in relation to the (P = 0.0038) in relation to k in the silence condition.
noise condition. LW was shorter for OLWJ (P = 0.0019),
MHSN (P = 0.0038) and GLS (P = 0.0038) than for the
DISCUSSION
silence condition.
The analysis results of the normalized clustering In the present study, we utilized PLI and then graph
coefficient c and the normalized characteristic path theoretical measures to investigate the effects of music
length k are shown in Table 4. Statistical differences on the neural networks of EEG, in different frequency
could be observed within the alpha2 band. The pair wise bands. We observed that functional connectivity, as
comparisons revealed lower c in the music conditions expressed by PLI, increased in music conditions
than in the noise condition: for OLWJ (P = 0.0038), compared with noise and silence conditions in the
MHSN (P = 0.0027) and GLS (P = 0.0045). c alpha2 band. Moreover, network analysis results
decreased for OLWJ (P = 0.0045), MHSN (P = 0.0019) demonstrated salient differences between music and
and GLS (P = 0.0038) in comparison with the silence noise conditions and between music and silence
condition. For k, the values showed lower during conditions in the alpha2 band: lower normalized

Table 3. Mean clustering coefficients CW and characteristic path lengths LW for five sound backgrounds in six frequency bands

Bands OLWJ MHSN GLS Noise Silence

Mean ± SD Mean ± SD Mean ± SD Mean ± SD Mean ± SD

CW
Delta 0.1471 ± 0.0211 0.1531 ± 0.0177 0.1560 ± 0.0291 0.1708 ± 0.0272 0.1825 ± 0.0236
Theta 0.1111 ± 0.0361 0.1408 ± 0.0511 0.1290 ± 0.0465 0.1073 ± 0.0290 0.0998 ± 0.0260
Alpha1 0.2449 ± 0.0737 0.2562 ± 0.0879 0.2471 ± 0.0786 0.1827 ± 0.0496 0.1712 ± 0.0461
Alpha2 0.2535a,b ± 0.0694 0.2620a,b ± 0.0846 0.2505a,b ± 0.0769 0.1772a ± 0.0491 0.1727b ± 0.0426
Beta 0.1131 ± 0.0357 0.1282 ± 0.0434 0.1234 ± 0.0379 0.0897 ± 0.0254 0.0865 ± 0.0231
Gamma 0.0431 ± 0.0067 0.0430 ± 0.0070 0.0409 ± 0.0057 0.0411 ± 0.0063 0.0402 ± 0.0088

LW
Delta 5.9712 ± 0.7179 5.6502 ± 0.5414 5.7658 ± 0.7500 5.2700 ± 0.7674 4.8995 ± 0.5453
Theta 10.1972 ± 2.0298 8.5594 ± 1.7975 9.5436 ± 2.1770 10.2188 ± 2.1734 10.5633 ± 2.0963
Alpha1 5.4992 ± 1.7259 5.7722 ± 2.0668 5.6632 ± 1.8426 6.5982 ± 1.6138 7.1106 ± 1.8020
Alpha2 5.0650a,b ± 1.5126 5.2506a,b ± 1.6926 5.7379a,b ± 1.5737 6.9551a ± 1.8377 6.8481b ± 1.6147
Beta 10.4645 ± 2.5420 9.6408 ± 2.2002 10.1846 ± 2.8192 12.0906 ± 1.8751 13.2112 ± 3.0676
Gamma 21.7616 ± 2.8352 21.7082 ± 3.0402 22.2068 ± 2.8199 22.4996 ± 2.9063 23.3341 ± 2.8873
For each band, letter superscripts indicate statistically significant differences between backgrounds (Wilcoxon signed rank test, P < 0.005, Bonferroni-corrected).
a
Significant differences between music and noise conditions.
b
Significant differences between music and silence conditions.

Table 4. Mean normalized parameters CW/CW-s and LW/LW-s for five sound backgrounds in six frequency bands

Bands OLWJ MHSN GLS Noise Silence

Mean ± SD Mean ± SD Mean ± SD Mean ± SD Mean ± SD

c = CW/CW-s
Delta 1.1329 ± 0.0255 1.1291 ± 0.0330 1.1135 ± 0.0234 1.1128 ± 0.0252 1.1592 ± 0.0297
Theta 1.1216 ± 0.0262 1.1090 ± 0.0272 1.1350 ± 0.0408 1.1139 ± 0.0282 1.1244 ± 0.0383
Alpha1 1.0961 ± 0.0325 1.0800 ± 0.0295 1.1240 ± 0.0460 1.1304 ± 0.0433 1.1525 ± 0.0479
Alpha2 1.0853a,b ± 0.0303 1.0696a,b ± 0.0292 1.0873a,b ± 0.0264 1.1529a ± 0.0517 1.1652b ± 0.0556
Beta 1.1375 ± 0.0359 1.1557 ± 0.0489 1.1568 ± 0.0472 1.1635 ± 0.0322 1.1685 ± 0.0618
Gamma 1.1909 ± 0.0560 1.1602 ± 0.0434 1.1836 ± 0.0596 1.1869 ± 0.0560 1.1679 ± 0.0602

k = LW/LW-s
Delta 1.0131 ± 0.0169 1.0147 ± 0.0209 1.0227 ± 0.0203 1.0139 ± 0.0200 1.0170 ± 0.0159
Theta 1.0621 ± 0.0257 1.0483 ± 0.0213 1.0727 ± 0.0358 1.0437 ± 0.0212 1.0455 ± 0.0233
Alpha1 1.0483 ± 0.0194 1.0346 ± 0.0204 1.0589 ± 0.0335 1.0618 ± 0.0193 1.0728 ± 0.0241
Alpha2 1.0396a,b ± 0.0190 1.0332a,b ± 0.0181 1.0420a,b ± 0.0195 1.0767a ± 0.0215 1.0833b ± 0.0266
Beta 1.0744 ± 0.0270 1.0760 ± 0.0248 1.0770 ± 0.0329 1.0837 ± 0.0268 1.0829 ± 0.0355
Gamma 1.0684 ± 0.0453 1.0524 ± 0.0405 1.0441 ± 0.0385 1.0329 ± 0.0318 1.0515 ± 0.0273
For each band, letter superscripts indicate statistically significant differences between backgrounds (Wilcoxon signed rank test, P < 0.005, Bonferroni-corrected).
a
Significant differences between music and noise conditions.
b
Significant differences between music and silence conditions.
 J. Wu et al. / Neuroscience 250 (2013) 49–59 57

clustering coefficient and characteristic path length during In the network analysis, the most salient results were
music listening. These variances reflect a tendency of the an increased CW and a decreased LW in the alpha2 band
functional networks of the brain toward a more random while listening to music as compared with the noise and
structure. silence conditions (Table 3). It is noteworthy that the
During the past decades, music perception has values of CW and LW depend on the level of PLI besides
attracted great attention with respect to its neural network organization. An enhancement of the mean
correlates by means of lesion studies and functional synchronization strength would give rise to a higher CW
imaging techniques. Most of these studies are primarily and a shorter LW. A detailed comparison of these
based on the notion of modularity for music processing results with a previous study (Wu et al., 2012), in which
(Peretz and Coltheart, 2003), where a module CW and LW were compared among music, noise and
corresponds to a specific brain area devoted to aspects silence conditions in the alpha frequency range for the
of music perception, including musical feature analysis, same threshold, should be undertaken. Wu et al. (2012)
auditory memory, auditory scene analysis, and reported that LW was smaller in music perception, with
processing of musical syntax and semantics (Koelsch relatively larger CW. Although the results of both studies
and Siebel, 2005; Koelsch, 2011). The neural correlates are in agreement, variances in CW and LW cannot be
of each aspect could be explored with laboratory music, simply deemed as genuine changes in network
composed by various manipulations of the organization since they are possibly ascribed to the
corresponding musical aspect (Peretz and Zatorre, increase of PLI for music listening.
2005; Ruiz et al., 2009; Schulze et al., 2011). However, To solve this problem, the normalized measures c and
in reality, music is an orderly sequence of musical k were calculated by comparing each original network
attributes, which possess different complicacies. The with its corresponding random networks. In this way,
present study thus chose the music heard in our daily both measures were corrected for the changes in mean
lives as auditory stimulus to reconstruct the natural PLI between background conditions. We found lower c
surroundings in the laboratory, and probed the neural and k during music listening in the alpha2 band (Table
correlates via functional co-ordination of cortical areas. 4), implying that the brain shifted toward a more random
The increased degree of synchronization in the alpha configuration for music perception. Note that the
band while listening to music (Table 2) has also been normalized clustering coefficient c was quite close to
reported in several EEG studies (Petsche et al., 1997; 1.0, which was much smaller than those of several EEG
Flores-Gutiérrez et al., 2009; Wu et al., 2012). Our and MEG studies (Stam et al., 2007a; Rubinov et al.,
finding provides further support to the notion that music 2009; Boersma et al., 2011; Douw et al., 2011). It is
perception, just like other higher cognitive function, probable that, owing to volume conduction, spurious
requires neural processes in multiple cortical areas interdependencies between recorded signals would
working in a co-operative manner (Bhattacharya and result in high evaluations of the clustering coefficients in
Petsche, 2005). Combining this finding with the results previous studies.
of power spectral analysis (Table 1), we inferred that Lower clustering c in the music backgrounds could be
the increased synchronizations under music perception explained by an economical principle (Bassett et al.,
were not biased by power differences since the 2010; Kitzbichler et al., 2011; Bullmore and Sporns,
differences in mean PSD among the sound 2012). Densely connected cortical areas or neurons,
backgrounds were not significant. characterized by high clustering, incline to be spatial
Music perception showed an increase of the neighbors and consequently reduce wiring cost.
intraregional and interregional synchronizations (Fig. 3), Therefore decreased clustering indicates higher wiring
suggesting that various cortical areas were implicated in cost of functional networks. In addition, path length k
music processing even though these areas were showed lower values in the music conditions, signifying
physically distant. Enhanced synchronous oscillations that the shortest routes between brain regions
among posterior regions may be related to the sensory decreased so as to improve parallel efficiency of
stage of the acoustic information processing in information transfer (Stam and Reijneveld, 2007;
association areas (Flores-Gutiérrez et al., 2009). The Bullmore and Sporns, 2012). The decrease in both c
involvement of the left temporal area implied that and k could be interpreted as a shift of network
musical stimuli activated brain regions which recruited in structure toward a more efficient but less economical
language processing. This corroborates the hypothesis configuration. The changes in brain organization in
of an intimate relation between music and language, response to music resemble those in working memory
indicating that overlapping neural resources are involved task performance (Kitzbichler et al., 2011). This
in both music and language appraisal (Koelsch and similarity suggests that the music listening task may
Siebel, 2005; Flores-Gutiérrez et al., 2007; Koelsch, demand greater information processing and cognitive
2011). Furthermore, functional interactions among task- effort.
relevant areas within the alpha band were strengthened One of the aims of the present research was to look
under music perception. In terms of the framework for possible differences in network organization between
proposed by Palva and Palva (2011), this finding offers musical excerpts. However, no differences could be
evidence for the assumption that alpha rhythmicity plays found between musical stimuli. These might be due to
a role in active task-relevant processing, instead of the the fact that the styles of the chosen musical excerpts
suppression of task-irrelevant processing. were similar.
58 J. Wu et al. / Neuroscience 250 (2013) 49–59

A potential criticism might be that the differences in Bassett DS, Meyer-Lindenberg A, Achard S, Duke T, Bullmore E
synchronization and network measures merely reflected (2006) Adaptive reconfiguration of fractal small-world human
brain functional networks. Proc Nat Acad Sci 103:19518–19523.
transient changes between stimuli. We discarded the
Bassett DS, Wymbs NF, Porter MA, Mucha PJ, Carlson JM, Grafton
first and last 5-s EEG records of each auditory epoch to ST (2011) Dynamic reconfiguration of human brain networks
avoid transition effects. Furthermore, if the transition during learning. Proc Nat Acad Sci 108:7641–7646.
effects had an impact, there should be distinct Bhattacharya J, Petsche H (2005) Phase synchrony analysis of EEG
differences between stimuli. But in fact it was not true. during music perception reveals changes in functional
The changes in PLI and network configuration connectivity due to musical expertise. Signal Process
accordingly could not be attributed to transient changes 85:2161–2177.
Bhattacharya J, Petsche H, Pereda E (2001) Interdependencies in
between stimuli. Also, little difference between the noise the spontaneous EEG while listening to music. Int J
and silence conditions further indicates that the task- Psychophysiol 42:287–301.
related functional reorganization is brought about by the Boersma M, Smit DJA, de Bie HMA, Van Baal GCM, Boomsma DI,
musical effect rather than the acoustic effect. de Geus EJC, Delemarre-van de Waal HA, Stam CJ (2011)
Another possible confusion could be the use of fixed Network analysis of resting state EEG in the developing young
frequency ranges. There may be individual variability in brain: structure comes with maturation. Hum Brain Mapp
32:413–425.
any certain frequency. But individually determined filter
Bullmore E, Sporns O (2009) Complex brain networks: graph
setting directly results in a comparison problem, and theoretical analysis of structural and functional systems. Nat
moreover, clear boundaries between relevant bands and Rev Neurosci 10:186–198.
their probable functional roles cannot be ascertained Bullmore E, Sporns O (2012) The economy of brain network
(Bhattacharya and Petsche, 2005). Hence, for simplistic organization. Nat Rev Neurosci 13:336–349.
account, we analyzed the EEG data by employing Bullmore ET, Bassett DS (2011) Brain graphs: graphical models of
the human brain connectome. Annu Rev Clin Psychol 7:113–140.
standard EEG frequency bands.
Deuker L, Bullmore ET, Smith M, Christensen S, Nathan PJ,
It is controversial whether different frequency effects Rockstroh B, Bassett DS (2009) Reproducibility of graph
exist when the alpha band is split into lower and upper metrics of human brain functional networks. NeuroImage
bands. While several studies have observed distinct 47:1460–1468.
reactivity patterns for the two alpha sub-bands, others Douw L, Schoonheim MM, Landi D, van der Meer ML, Geurts JJG,
have revealed similar patterns (Fink et al., 2005; Reijneveld JC, Klein M, Stam CJ (2011) Cognition is related to
Micheloyannis et al., 2009). There is empirical evidence resting-state small-world network topology: an
magnetoencephalographic study. Neuroscience 175:169–177.
that functional specialization of different alpha frequency Eguı́luz VM, Chialvo DR, Cecchi GA, Baliki M, Apkarian AV (2005)
bands emerges with increasing task demands (Fink Scale-free brain functional networks. Phys Rev Lett 94:018102.
et al., 2005). In this study, the results for both lower and Fink A, Grabner RH, Neuper C, Neubauer AC (2005) EEG alpha
upper alpha bands were not identical. band dissociation with increasing task demands. Cogn Brain Res
24:252–259.
Flores-Gutiérrez EO, Dı́az J-L, Barrios FA, Favila-Humara R,
CONCLUSION Guevara MÁ, del Rı́o-Portilla Y, Corsi-Cabrera M (2007)
Metabolic and electric brain patterns during pleasant and
In summary, our study provides evidence for the
unpleasant emotions induced by music masterpieces. Int J
reconfiguration of human brain functional networks while Psychophysiol 65:69–84.
listening to music. We found that music perception Flores-Gutiérrez EO, Dı́az J-L, Barrios FA, Guevara MÁ, del Rı́o-
induced an increased synchronization of cortical regions Portilla Y, Corsi-Cabrera M, del Flores-Gutiérrez EO (2009)
and a more random network configuration. The pattern Differential alpha coherence hemispheric patterns in men and
changes could be detected particularly in the upper women during pleasant and unpleasant musical emotions. Int J
Psychophysiol 71:43–49.
alpha band. Furthermore, differences in network
Fries P (2005) A mechanism for cognitive dynamics: neuronal
architecture were not observed between musical stimuli. communication through neuronal coherence. Trends Cogn Sci
These findings imply that music perception leads to an 9:474–480.
emergence of a more efficient but less economical Gray PM, Krause B, Atema J, Payne R, Krumhansl C, Baptista L
structure and requires more information processing as (2001) The music of nature and the nature of music. Science
well as cognitive effort. 291:52–54.
Gross J, Kujala J, Hämäläinen M, Timmermann L, Schnitzler A,
Salmelin R (2001) Dynamic imaging of coherent sources:
Acknowledgments—The work described in this paper was sup- studying neural interactions in the human brain. Proc Nat Acad
ported by National Basic Research Program of China (Grant Sci 98:694–699.
No. 2013CB329502) and National Nature Science Foundation Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen
of China (Grant Nos. 61273338 and 31200769). VJ, Sporns O (2008) Mapping the structural core of human
cerebral cortex. PLoS Biol 6:e159.
Hayasaka S, Laurienti PJ (2010) Comparison of characteristics
REFERENCES between region- and voxel-based network analyses in resting-
state fMRI data. NeuroImage 50:499–508.
He Y, Chen ZJ, Evans AC (2007) Small-world anatomical networks in
Amor F, Baillet S, Navarro V, Adam C, Martinerie J, Le Van Quyen M
the human brain revealed by cortical thickness from MRI. Cereb
(2009) Cortical local and long-range synchronization interplay in
Cortex 17:2407–2419.
human absence seizure initiation. NeuroImage 45:950–962.
He Y, Evans A (2010) Graph theoretical modeling of brain
Bassett DS, Greenfield DL, Meyer-Lindenberg A, Weinberger DR,
connectivity. Curr Opin Neurol 23:341–350.
Moore SW, Bullmore ET (2010) Efficient physical embedding of
He Y, Wang J, Wang L, Chen ZJ, Yan C, Yang H, Tang H, Zhu C,
topologically complex information processing networks in brains
Gong Q, Zang Y, Evans AC (2009) Uncovering intrinsic modular
and computer circuits. PLoS Comput Biol 6:e1000748.
 J. Wu et al. / Neuroscience 250 (2013) 49–59 59

organization of spontaneous brain activity in humans. PLoS ONE Petsche H, Kaplan S, von Stein A, Filz O (1997) The possible
4:e5226. meaning of the upper and lower alpha frequency ranges for
Iturria-Medina Y, Sotero RC, Canales-Rodrı́guez EJ, Alemán-Gómez cognitive and creative tasks. Int J Psychophysiol 26:77–97.
Y, Melie-Garcı́a L (2008) Studying the human brain anatomical Rubinov M, Knock SA, Stam CJ, Micheloyannis S, Harris AWF,
network via diffusion-weighted MRI and graph theory. Williams LM, Breakspear M (2009) Small-world properties of
NeuroImage 40:1064–1076. nonlinear brain activity in schizophrenia. Hum Brain Mapp
Jin S-H, Lin P, Hallett M (2012) Reorganization of brain functional 30:403–416.
small-world networks during finger movements. Hum Brain Mapp Ruiz MH, Koelsch S, Bhattacharya J (2009) Decrease in early right
33:861–872. alpha band phase synchronization and late gamma band
Kay BP, Meng X, DiFrancesco MW, Holland SK, Szaflarski JP (2012) oscillations in processing syntax in music. Hum Brain Mapp
Moderating effects of music on resting state networks. Brain Res 30:1207–1225.
1447:53–64. Schmithorst VJ (2005) Separate cortical networks involved in music
Kitzbichler MG, Henson RNA, Smith ML, Nathan PJ, Bullmore ET perception: preliminary functional MRI evidence for modularity of
(2011) Cognitive effort drives workspace configuration of human music processing. NeuroImage 25:444–451.
brain functional networks. J Neurosci 31:8259–8270. Schnitzler A, Gross J (2005) Normal and pathological oscillatory
Kitzbichler MG, Smith ML, Christensen SR, Bullmore E (2009) communication in the brain. Nat Rev Neurosci 6:285–296.
Broadband criticality of human brain network synchronization. Schulze K, Mueller K, Koelsch S (2011) Neural correlates of strategy
PLoS Comput Biol 5:e1000314. use during auditory working memory in musicians and non-
Koelsch S (2011) Towards a neural basis of music perception – a musicians. Eur J Neurosci 33:189–196.
review and updated model. Front Psychol 2:110. Semlitsch HV, Anderer P, Schuster P, Presslich O (1986) A solution
Koelsch S, Siebel WA (2005) Towards a neural basis of music for reliable and valid reduction of ocular artifacts, applied to the
perception. Trends Cogn Sci 9:578–584. P300 ERP. Psychophysiology 23:695–703.
Latora V, Marchiori M (2001) Efficient behavior of small-world Sporns O, Honey CJ, Kötter R (2007) Identification and classification
networks. Phys Rev Lett 87:198701. of hubs in brain networks. PLoS ONE 2:e1049.
Lehmann D, Faber PL, Gianotti LRR, Kochi K, Pascual-Marqui RD Sporns O, Tononi G, Edelman GM (2000) Connectivity and
(2006) Coherence and phase locking in the scalp EEG and complexity: the relationship between neuroanatomy and brain
between LORETA model sources, and microstates as putative dynamics. Neural Netw 13:909–922.
mechanisms of brain temporo-spatial functional organization. J Sporns O, Zwi J (2004) The small world of the cerebral cortex.
Physiol Paris 99:29–36. Neuroinformatics 2:145–162.
Micheloyannis S, Vourkas M, Tsirka V, Karakonstantaki E, Kanatsouli Stam CJ, de Haan W, Daffertshofer A, Jones BF, Manshanden I, van
K, Stam CJ (2009) The influence of ageing on complex brain Cappellen van Walsum AM, Montez T, Verbunt JPA, de Munck
networks: a graph theoretical analysis. Hum Brain Mapp JC, van Dijk BW, Berendse HW, Scheltens P (2009) Graph
30:200–208. theoretical analysis of magnetoencephalographic functional
Newman MEJ (2003) The structure and function of complex connectivity in Alzheimer’s disease. Brain 132:213–224.
networks. SIAM Rev 45:167–256. Stam CJ, Jones BF, Nolte G, Breakspear M, Scheltens P (2007a)
Nolte G, Bai O, Wheaton L, Mari Z, Vorbach S, Hallett M (2004) Small-world networks and functional connectivity in Alzheimer’s
Identifying true brain interaction from EEG data using the disease. Cereb Cortex 17:92–99.
imaginary part of coherency. Clin Neurophysiol 115:2292–2307. Stam CJ, Nolte G, Daffertshofer A (2007b) Phase lag index:
Nunez PL, Srinivasan R, Westdorp AF, Wijesinghe RS, Tucker DM, assessment of functional connectivity from multi channel EEG
Silberstein RB, Cadusch PJ (1997) EEG coherency: I: statistics, and MEG with diminished bias from common sources. Hum Brain
reference electrode, volume conduction, Laplacians, cortical Mapp 28:1178–1193.
imaging, and interpretation at multiple scales. Stam CJ, Reijneveld JC (2007) Graph theoretical analysis of complex
Electroencephalogr Clin Neurophysiol 103:499–515. networks in the brain. Nonlinear Biomed Phys 1:3.
Palva S, Monto S, Palva JM (2010) Graph properties of synchronized Tian L, Wang J, Yan C, He Y (2011) Hemisphere- and gender-related
cortical networks during visual working memory maintenance. differences in small-world brain networks: a resting-state
NeuroImage 49:3257–3268. functional MRI study. NeuroImage 54:191–202.
Palva S, Palva JM (2011) Functional roles of alpha-band phase Watts DJ, Strogatz SH (1998) Collective dynamics of ‘small-world’
synchronization in local and large-scale cortical networks. Front networks. Nature 393:440–442.
Psychol 2:204. Wu J, Zhang J, Liu C, Liu D, Ding X, Zhou C (2012) Graph theoretical
Pearce MT, Ruiz MH, Kapasi S, Wiggins GA, Bhattacharya J (2010) analysis of EEG functional connectivity during music perception.
Unsupervised statistical learning underpins computational, Brain Res 1483:71–81.
behavioural, and neural manifestations of musical expectation. Yan C, Gong G, Wang J, Wang D, Liu D, Zhu C, Chen ZJ, Evans A,
NeuroImage 50:302–313. Zang Y, He Y (2011) Sex- and brain size-related small-world
Peretz I, Coltheart M (2003) Modularity of music processing. Nat structural cortical networks in young adults: a DTI tractography
Neurosci 6:688–691. study. Cereb Cortex 21:449–458.
Peretz I, Zatorre RJ (2005) Brain organization for music processing. Zhu W, Zhang J, Ding X, Zhou C, Ma Y, Xu D (2009) Crossmodal
Annu Rev Psychol 56:89–114. effects of Guqin and piano music on selective attention: an event-
Petermann T, Thiagarajan TC, Lebedev MA, Nicolelis MAL, Chialvo related potential study. Neurosci Lett 466:21–26.
DR, Plenz D (2009) Spontaneous cortical activity in awake Zhu W, Zhang J, Liu H, Ding X, Ma Y, Zhou C (2008) Differential
monkeys composed of neuronal avalanches. Proc Nat Acad Sci cognitive responses to guqin music and piano music in Chinese
106:15921–15926. subjects: an event-related potential study. Neurosci Bull 24:21–28.

(Accepted 13 June 2013)

(Available online 24 June 2013)