Vous êtes sur la page 1sur 12

Computers in Biology and Medicine 67 (2015) 4960

Contents lists available at ScienceDirect

Computers in Biology and Medicine


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

Nonlinear analysis of EEGs of patients with major depression during


different emotional states
Saime Akdemir Akar a,n, Sadk Kara a, Smeyra Agambayev a, Vedat Bilgi b
a
b

Institute of Biomedical Engineering, Fatih University, Istanbul 34500, Turkey


Department of Psychiatry, School of Medicine, Fatih University, Istanbul 34500, Turkey

art ic l e i nf o

a b s t r a c t

Article history:
Received 9 July 2015
Accepted 24 September 2015

Background: Although patients with major depressive disorder (MDD) have dysfunctions in cognitive
behaviors and the regulation of emotions, the underlying brain dynamics of the pathophysiology are
unclear. Therefore, nonlinear techniques can be used to understand the dynamic behavior of the EEG
signals of MDD patients.
Methods: To investigate and clarify the dynamics of MDD patients' brains during different emotional
states, EEG recordings were analyzed using nonlinear techniques. The purpose of the present study was
to assess whether there are different EEG complexities that discriminate between MDD patients and
healthy controls during emotional processing. Therefore, nonlinear parameters, such as Katz fractal
dimension (KFD), Higuchi fractal dimension (HFD), Shannon entropy (ShEn), Lempel-Ziv complexity
(LZC) and Kolmogorov complexity (KC), were computed from the EEG signals of two groups under different experimental states: noise (negative emotional content) and music (positive emotional content)
periods.
Results: First, higher complexity values were generated by MDD patients relative to controls. Signicant
differences were obtained in the frontal and parietal scalp locations using KFD (po0.001), HFD
(po 0.05), and LZC (p 0.05). Second, lower complexities were observed only in the controls when they
were subjected to music compared to the resting baseline state in the frontal (p o0.05) and parietal
(p 0.005) regions. In contrast, the LZC and KFD values of patients increased in the music period compared to the resting state in the frontal region (po0.05). Third, the patients' brains had higher complexities when they were exposed to noise stimulus than did the controls' brains. Moreover, MDD
patients' negative emotional bias was demonstrated by their higher brain complexities during the noise
period than the music stimulus. Additionally, we found that the KFD, HFD and LZC values were more
sensitive in discriminating between patients and controls than the ShEn and KC measures, according to
the results of ANOVA and ROC calculations.
Conclusion: It can be concluded that the nonlinear analysis may be a useful and discriminative tool in investigating the neuro-dynamic properties of the brain in patients with MDD during emotional stimulation.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
EEG
Major depressive disorder
Complexity
Emotion
Music
Noise
Entropy
Fractal dimension

1. Introduction
Major depressive disorder (MDD) is a very common psychiatric
mood disorder that affects 1520% of the population [1]. The major
signs of MDD are loss of interest, energy and pleasure; a depressed
mood; disturbances in sleep; and recurrent suicidal thoughts [2].
Moreover, depressed patients exhibit some symptoms related to
cognition, such as low concentration, difculty in decision making
and focusing, according to the Diagnostic and Statistical Manual of
Mental Disorders (DSM-IV) diagnostic criteria for MDD [3].

Corresponding author. Tel.: 90 212 8663300x2643; fax: 90 212 8663412.


E-mail address: saimeakar@fatih.edu.tr (S. Akdemir Akar).

http://dx.doi.org/10.1016/j.compbiomed.2015.09.019
0010-4825/& 2015 Elsevier Ltd. All rights reserved.

However, the underlying neural activities of these types of functional or cognitive impairments are still under investigation [4].
Therefore, there have been many electroencephalogram (EEG)
studies in groups of patients diagnosed with MDD to compare
them with healthy controls. Most of these studies have investigated resting state brain activity in MDD patients with the help of
conventional linear techniques, such as spectral domain analysis
or waveform investigations [5-8]. However, the complex dynamic
variations in an EEG time series cannot be discriminated by linear
techniques [9]. Moreover, because EEG is a complex and irregular
signal with nonlinear behaviors and without a linear relation
between cause and effect, a signicant number of complexity
estimators, such as largest Lyapunov exponents (L1), correlation
dimension (D2), mutual information, Shannon entropy (ShEn),

50

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

sample entropy, approximate entropy (ApEn), Lempel-Ziv complexity (LZC), fractal dimension (FD), and Kolmogorov complexity
(KC), have been proposed as more appropriate techniques for
understanding the underlying dynamics of brain activity with
unknown parameters [4]. These estimators are known as nonlinear measures because they reconstruct an attractor from an EEG
by characterizing its dynamic behaviors using the dimension of a
signal, which gives the degrees of freedom of system, or using
entropies that reect the unpredictability of the signals' dynamics
[10].Therefore, over the past 40 years, these measures have been
applied to EEG data for characterizing random-appearing series of
patterns across time in different physiological states, such as Alzheimer's disease [11-13], dementia [14], epilepsy [15-17], schizophrenia [18-20] and obsessivecompulsive disorder [21].
It must be taken into account that all of these estimators reect
the complexity with different approaches, such as dimensional
complexity, regularity and predictability [22,23]. While the D2 and
Higuchi FD are related to dimensional complexity, the ApEn is a
statistical calculation of regularity of a signal. Some other complexity measures, such as the L1 and entropies, focus on determining the irregularity and predictability of signals. The randomness or degree of irregularity (chaoticness) of nite sequences can
be investigated by the LZC and KC. Among all of these complexity
estimators, traditional estimators (D2 and L1) are the oldest and
have limitations because of their requirement for a noise-free, long
and stationary time series [24]. Alternatively, entropy-based
complexity estimators calculated in either frequency domain
(spectral entropy) or a time series (ShEn, ApEn) have the advantage of anti-noise ability [25]. On the other hand, Katz fractal
dimension (KFD), Higuchi's fractal dimension (HFD) and Petrosian
FD algorithms simply and quickly estimate the self-similarity of a
time interval directly in the time series [26].
Although most of the attention has been paid to analyses of
EEGs in MDD patients based on linear techniques [5-8], there have
been a small number of studies on nonlinear EEG analysis of
patients [4,19,27-29]. To the best of our knowledge, Nandrino et al.
[27] conducted the rst nonlinear EEG analysis study of MDD
patients. They calculated the D2 and found a negative correlation
between the symptoms of MDD and its complexity. Li et al. [19]
reported that MDD patients exhibited increased LZC values in
anterior brain regions when the values were compared with corresponding values in controls. Ahmadlou et al. [28] applied HFD
and KFD analyses to both the full band and sub-bands of EEG data
in patients with depression and in controls. According to their
reports, elevated complexities in patients were particularly
observed in the beta and gamma sub-bands of the frontal area in
EEG recordings. The same authors proposed a new nonlinear
measure called spatiotemporal analysis of relative convergence
(STARC) for investigating brain dynamics based on relative EEG
convergence in different loci [29]. They found signicant STARC
differences between male and female MDD patients. Puthankattil
and Joseph [30] used signal entropy values to classify the EEG
signals of patients with depression and control subjects. In another
of their works, the ApEn values of MDD patients and healthy
controls were compared [31]. They reported decreased ApEn
values, indicating the predictability and regularity of EEG recordings in patients with depression. In another study, Hosseinifard
et al. [32] compared the performance of different classication
algorithms that used some EEG nonlinear features, such as
detrended uctuation analysis (DFA), HFD, D2, and L1 for discriminating MDD patients from controls. They reported high
classication accuracy when the linear discriminant analysis was
used with extracted nonlinear features. Similarly, probabilistic
neural network-based high classication accuracy in the discrimination of MDD patients and control subjects was obtained by
Faust et al. [33]. They used entropy values estimated from the EEG

signals of two groups. To compare the depression-detecting ability


of linear and nonlinear EEG features, Bachmann et al. [34] calculated the HFD values of EEG signals and found high discrimination
accuracy when nonlinear methods were used. To show complex
dynamic variations in the EEG signals, computer-aided diagnosis
of depression using EEG signals has been proposed for psychiatrists as a tool to conrm their clinical diagnosis [9]. Finally, in a
very recent study, Bachmann et al. [4] investigated the LZC values
in MDD patients. Their results revealed increased complexities in
MDD patients compared to normal controls in all of the investigated brain regions.
Although all of these studies imply that nonlinear methods can
be useful for analyzing EEG signals in patients with depression, the
reported complexity results are restricted only to the analysis of
resting EEG recordings. Complexity approaches have not yet been
applied to the EEG signals of MDD patients during emotional
processing. It is known that depressed patients have impairments
in cognitive functions and the regulation of emotions [35]. However, the underlying factors of the relationship between depression and these impairments are largely unknown. This issue has
been investigated in previous studies that explored facial emotion
recognitions [35-37]. The processing of emotional stimuli was
found to be negatively biased in MDD patients, meaning that
patients focus more on negative facial expressions than on neutral
or positive expressions and produce more rapid and prominent
responses during negative ones [35-38]. These ndings are consistent with the results of neuroimaging studies that showed
decreased brain activation (in ventral striatum and putamen) to
positive stimuli in MDD patients relative to control subjects [3941]. In MDD patients, attenuated emotional reactivity to positive
emotional stimuli is called positive attenuation hypothesis, and
increased emotional reactivity to negative emotional stimuli is
known as negative potentiation hypothesis [42]. Most of these
previous studies investigated emotional processing in MDD
patients using neuroimaging ndings or self-report evaluations.
However, it has been reported that temporal dynamics play an
important role in understanding the neural structure of emotional
processing [43]. To evaluate the impairments in emotional
responses of MDD patients in the present study, which is the rst
of this kind to our knowledge, EEG data were recorded and analyzed by nonlinear techniques during both rest and different
emotional auditory stimulation periods, such as listening to classical Turkish music (CTM) and listening to noise. The primary goals
of this research were to determine whether complexity measures,
such as ApEn, ShEn, KFD, HFD and KC, can be used to discriminate
MDD patients from control subjects and whether the emotional
processing in the EEG signals can be measured by complexity
differences between patients and controls. The originality of this
research was to investigate several complexity and entropy features in the EEG signals, obtained from different brain regions, of
MDD patients during different emotional states. We designed a
new study protocol to test the hypothesis that MDD patients have
differences in brain complexities during different emotional stimulations and show a bias toward negative emotional stimuli. In
particular, we hypothesized that patients would have increased
brain complexities both at rest and during noise stimulation
periods.

2. Methods
2.1. Participants and experimental protocol
The groups in this study were healthy control subjects and
MDD patients. There were 15 control subjects, who had no history
of any neurological, psychiatric, or psychological disorder or

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

2.2. Data analysis

Table 1
Demographical and clinical features of participants.

N
Gender (male/female)
Age (years 7 std. deviation)
Education (years)
Age of onset
Baseline HAM-D Score

51

Healthy control
subjects

Patients with major


depression

15
8/7
29.42 7 4.02
15.337 4.02

15
8/7
30.92 7 3.65
13.92 7 2.78
28.127 3.63
22.147 3.98

HAM-D: Hamilton Rating Scale for Depression

substance abuse. There were also 15 age- and gender-matched


MDD patients diagnosed by an expert psychiatrist according to the
DSM-IV criteria. The degree of their symptoms was assessed by the
17-item Hamilton Rating Scale for Depression (HAM-D). The MDD
patients were being treated with a variety of medications (beta
blockers and other standard drugs) at the time of the experiments.
All participants reported that they were right-handed, had hearing
in the normal range, and had no previous formal music training
(Table 1).
The experimental protocol was approved by the Institutional
Ethical Board of Fatih University. All participants were aware of the
purpose and details of the study and gave their written informed
consent before the experiments. The EEG signals were recorded in
the Psychiatry Department of Sema Hospital at Fatih University.
The individuals were instructed to close their eyes, sit comfortably
in a chair without moving, breathe normally and suppress blinking
as much as possible. The experiments began after a pre-adaptation
period of 2 min, in which patients sat on a chair in a quiet and dark
room. The EEG data were acquired from 16 electrodes (Fp1, Fp2,
Fpz, F3, F4, F7, F8, C3, C4, Cz, P3, P4, P7, P8, T7, and T8) mounted on
the subject's scalp according to the International 1020 system.
The EEG signals were recorded using a BrainAmp DC acquisition
system (Brain Products, Munich, Germany) and an electrode cap
(Easy cap), and they were sampled with 250 Hz. Impedance values
of all electrodes were checked by online monitoring in the system
software to adjust values under 10 k. The experimental protocol
consisted of ve sequential periods for a total data recording
duration of 15 min. First, EEG signals were recorded from each
subject for a 3 min resting state (resting 1, R1). Second, during
another 3 min period, the EEGs were collected from subjects listening to music (M). Third, the data acquisition continued with a
second resting (R2) period of 3 min. In the following periods, data
were recorded from participants during exposure to noise (N) and
then during the third resting period (R3). Each participant was
instructed to listen to auditory stimuli via earphones with 75 dB
intensity. A special form/mode (Rast makam) of CTM was chosen
for the sedative musical stimulation period. This music was
instrumental and had been shown to induce cheerfulness, happiness and pleasure. The effects of different modes of Turkish music
on the emotions were dened by the Turkish philosopher Farabi
[44,45]. The sound of a broken washing machine was selected for
the alerting and disturbing noise stimulation period. Moreover, in
this study, subjective ratings of participants in terms of the emotions induced by noise and music stimuli were obtained for
standardization of the stimuli. After the experimental protocol,
questionnaires were given to all subjects. The participants rated
the auditory stimulation periods on a 15 scale for valence, with
1 being very negative or disturbing and 5 being very positive or
sedative. MDD patients rated the noise period (mean 7std 1.13 7 0.35) as more negative than the music period
(mean7 std 4.13 70.74). Similarly, the control subjects found the
noise period (mean 7std 1.27 70.46) to be more negative than
the music period (mean 7std 4.47 7 0.64).

In this study, we investigated the EEG data of participants using


various complexity measures, such as ShEn, LZC, KC, KFD and HFD,
using MATLAB (The MathWorks, Natick, MA, USA). The general
block diagram of the procedure is shown in Fig. 1. The data were
rst ltered using a 0.550 Hz band-pass lter for preprocessing.
Then, any artifact sources in terms of eye-blink and movement in
each EEG recording were eliminated automatically by the
recording system interface and by visual inspection by the expert.
Then, based on the results of previous studies [46], EEG data in
each period (180 s) were divided into 20 second segments, giving
9 segments for nonlinear processing. Next, the complexity measures were calculated for each segment. For calculating the LZC
and KC, each EEG data segment was converted into new binary
sequences by median thresholding. After that, the average values
of each complexity measure over all segments were computed for
each subject at each EEG electrode. Finally, the average complexity
values of the frontal, central, temporal and parietal regions were
calculated for each group and the differences between MDD
patients and controls in mean complexity measure during the
experimental protocol were statistically evaluated.
2.2.1. Shannon entropy
The rst denition of entropy was developed by Shannon and
Weaver for information theory [47]. It is a time domain complexity
estimator and can be calculated using the set of nite discrete

Start the procedure

EEG data acquisition in


each experimental period
(resting, music, and noise)

Pre-processing (noise and


artifact elimination)

Nonlinear feature extraction


Shannon entropy, Lempel-Ziv
complexity, Kolmogorov
complexity, Katzs fractal
dimension, Higuchis fractal
dimension

Statistical analysis
Fig. 1. The general block diagram of the procedure.

52

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

random variables, X fx1 ; x2 ; ; xm g,xi A Rd , as follows [48]


H X  c

m
X

pxi lnpxi

i0

where c shows a positive constant, m is the possible amplitude


values number, and pxi shows the probability of xi signal.

complexity measure is increased by one unit. Therefore, the LZC


shows the number of diverse patterns in a n-length binary
sequence and the nal complexity of the sequence bn can be
computed as
lim cn bn

n-1

where
2.2.2. Katz fractal dimension
The FD of a signal can be calculated using Katz's algorithm in
the time domain. This method depends on a simple calculation of
the FD of a planar curve as follows [49]
FD

log L
log d

where L sumdista; b and shows the total curve length or sum


of distances between successive points (a; b) in the curve (time
series), and d indicates the diameter. The average KFD can be
calculated using the following formula
FD

log L=a
log n

log d=a log n log d=L

2.2.3. Higuchi fractal dimension


Higuchi's algorithm estimates the FD of signals directly in the
time domain. X 1; X 2; ; X N shows the EEG time sequence
with N total samples. The algorithm sets up a new time series from
this given one as follows



 
N m
k
m 1; 2; k
xkm xm; xm k; xm 2k; x m
k
4
where m is the initial time, and k is the interval time between
successive points in the xkm sequence. The average normalized sum
of absolute values of each sequence can be calculated as

!
(
)

N1
XN  m

k
Lm k
5

xm ik xm i 1k
N  m
i1

k
k
where NNm1k is the normalization factor for the curve length of the
k

time series. Therefore, the average curve length for k sets of Lm k


can be calculated as
k
X

Lm k

m1

The FD value of Higuchi's method is then computed by a linear


D
tting procedure based on the least squares as Lk p k , which
indicates that the curve is fractal with D dimension [50].
2.2.4. Lempel-Ziv complexity
The LZC is a non-parametric estimator to assess the randomness of nite sequences [51]. To calculate the LZCs, time domain
data are converted into a binary sequence (0 and 1 values) as
follows
S s1; s2; ; sr
where s shows the binary sequence and is obtained by
(
0; xi o T d
si
1; otherwise

n
log 2n

10

To overcome the problem in length dependence of time series,


the LZC can be normalized by bn as [52]
LZC cn=bn

11

2.2.5. Kolmogorov complexity


Suggested by Kolmogorov [53] and Chaitin [54], KC, or algorithmic complexity, can be used for predicting the randomness or
degree of irregularity of symbolic sequences. The KC can be calculated as [55]
K U x min lp

where n aL shows the step's number in the curve or the time


series.

Lk

bn

p:U p x

12

where p indicates the computer program, and lp shows the


length of output string (x) of the U universal Turing machine
(computer). For calculating the KC of a time series, rst, data must
be transformed into discrete binary sequences as in LZC.
2.3. Statistical analysis
Statistical data analysis was performed using SPSS 20.0
s
software (SPSS Inc., Chicago, Illinois, USA). First, the data were
checked for normality and homogeneity of variance using the
ShapiroWilks test. A three-way repeated measures (mixed) analysis of variance (ANOVA) was used to investigate signicant group
differences in each of the complexity measures for the different
experimental periods (R1, M, R2, N, R3) and four brain regions
(frontal, central, temporal and parietal) and between the patients
and controls. The within-group factors were the experimental
periods and electrode locations, and the between-group factor was
the groups (the depressed and the control groups). To control
Type-1 error of F ratios, GreenhouseGeisser corrections were also
performed. Bonferroni post-hoc tests were performed to assess the
sources of signicant results of ANOVA. When a signicant group x
region or group x region x emotional period interaction were
detected in each complexity value by mixed ANOVA, a one-way
ANOVA was performed to clarify the signicance of difference
between patients and controls for each of the regions. Therefore,
the most distinctive complexity values and corresponding brain
areas and emotional states between the MDD patients and control
subjects were assessed. To investigate how these complexity
measures change from the resting baseline period to the emotional condition related to noise and music stimuli in each group,
paired sample Student's t-tests were used by pairing each set of
sequential periods. The statistical analysis results were accepted as
signicant when the p value was less than 0.05.

7
3. Results
8

The xi time series value is dened as one if it is greater than a


threshold (T d ) and as zero if it is lower than or equal to T d [19].
After this transformation process, the binary sequence si is
scanned from left to right to count the number of new occurring
patterns. When a new subsequence pattern is observed, the cn

In this study, average EEG signal complexity values (KFD, HFD,


ShEn, LZC and KC) of MDD patients and healthy controls were
investigated during different emotional states for channels in the
frontal (Fp1, Fp2, Fpz, F3, F4, F7, F8), central (C3, Cz, C4), temporal
(T7, T8) and parietal (P3, P4, P7, P8) regions. To evoke an emotional
condition, two different auditory stimuli (noise and music) were
used, and their effects on nonlinear brain dynamics of patients and

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

controls were evaluated. A remarkable change in EEG complexity


values between patients and controls was observed in association
with emotional stimulation.

53

po 0.001] and ShEn [F(3,51) 3.981, p 0.013] values. Finally,


signicant region x group x period interactions were observed in
KFD [F(12, 204) 2.098, p 0.018] and LZC [F(12, 204) 1.876,
p 0.039] values.

3.1. Group differences


Signicant complexity (KFD, HFD, and LZC) differences
between patients and controls were obtained in the resting
baseline period (R1). Consistent with the hypothesis, patients with
MDD had higher EEG complexities than control subjects during all
periods of the procedure, as shown in Figs. 24. Repeated measures ANOVA exhibited signicant main effects of group in KFD [F
(1,17) 40.667, p o0.001], HFD [F(1,17) 12.730, p 0.002], and
LZC [F(1,17) 6.791, p 0.018] values. To understand which analyses have a signicant effect, p-values obtained from one-way
ANOVA for each complexity measure for each stimulation period
are listed in Tables 2 and 3. p-values of each complexity measure
in any period lower than those in the other periods indicate the
most distinctive complexity values and corresponding brain areas
and emotional states between the MDD and control groups. The
signicant differences between the two groups were obtained
mostly in the parietal and frontal scalp areas by the KFD, HFD and
LZC measures. In the rst baseline (R1) period, it was found that
the complexity values obtained by the KFD, HFD and LZC measures
of patients' frontal and parietal brain regions were higher than
those of controls.
A three-way repeated measures ANOVA showed that signicant brain region effects were evident in all complexity measures as KFD [F(3,51) 14.089, p 0.002], HFD [F(3,51) 112.144,
p 0.001], ShEn [F(3,51) 3.702, p 0.017], LZC [F(3,51) 2.874,
p 0.045] and KC [F(3,51) 3.580, p 0.020] values. Post-hoc
comparisons indicated that the KFD values of the frontal and
parietal regions were signicantly higher than the values of other
areas. Similarly, the HFD, ShEn, LZC and KC values found to be
higher in the frontal area than in the other three regions. The
group x region interactions in complexity values were also signicant in KFD [F(3,51) 38.446, p o0.001], HFD [F(3,51) 13.533,

3.2. Changes in complexity measures from resting to emotional


periods
The control subjects showed reduced complexity values especially in the frontal and parietal regions, elicited by M stimulus.
Namely, this signicant difference in the KFD measure from resting state to M stimulus was obtained in both the frontal (p 0.039)
and parietal (p 0.005) scalp regions. Moreover, the HFD

Fig. 3. Comparison of KFD values during each experimental period (Abbreviations:


R1 resting 1, M musical stimulus, R2 resting 2, N noise stimulus, R3 resting
3) between control and MDD groups in frontal, central, parietal and temporal
regions. * and ** indicate the statistical differences between groups of po 0.05 and
p o0.001, respectively.

Fig. 2. Comparison of LZC values using bar-graph representation during each experimental period (Abbreviations: R1 resting 1, M musical stimulus, R2 resting 2,
N noise stimulus, R3 resting 3) between control and MDD groups in frontal, central, parietal and temporal regions. * and ** indicate the statistical differences between
groups of p o0.05 and p o0.001, respectively.

54

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

complexity values in the MDD group were not signicantly lower


in the M period compared to the baseline period over the parietal
region. Interestingly, the MDD group displayed the opposite trend
as the LZC and KFD values increased in the M period compared to
the rst baseline state (R1) over the frontal region (p o0.05). In
contrast, the results indicated that during exposure to the noise,
the randomness or the complexity by KFD in the EEG signals of
MDD patients was increased in the frontal (po0.001) and parietal
(p o0.05) regions. This signicant difference was not present for
the central and temporal regions. Moreover, the LZC values of
MDD patients increased from baseline to noise period in the parietal region. Similarly, according to the paired sample t-test results,
the KFD and LZC values during the N period were signicantly
higher than those of the resting states in the control group
(p o0.05). However, the amount of increase in the groups was
different. The results showed that the MDD group had the largest
overall KFD increase during the noise period compared to the
resting baseline state in the frontal brain region. There was not a
statistically signicant difference in the complexity values
between the two groups during the resting period (R3) after the
stimulation periods, except in the LZC values of the frontal region.
The complexity values generally tended to revert to the levels
observed before the stimulation periods in both groups. The EEG
complexities decreased, especially in the frontal (by KFD) and
parietal (by HFD) regions, when the results of the R1 and R3
periods were compared.
Overall, we also found that the LZC values of the MDD and
control groups were closer after the stimulation periods (R3) than

at rst baseline (R1). This nding might be related to increased


complexities in control subjects during noise stimulus. Moreover,
according to the paired sample t-test results, LZC values decreased
from noise to the R3 period. Therefore, the difference between
groups diminished when looking at R3 compared to R1. ANOVA
exhibited a signicant effect of periods in the KFD values [F
(4,68) 11.866, p 0.001]. The highest KFD values were obtained
in the N stimulation period. ANOVA revealed signicant results for
the stimulation period x group interaction in KFD [F(4,68) 5.974,
po 0.001] and LZC [F(4,68) 5.443, p 0.001] values. Multiple
comparisons showed that while signicant differences in KFD
values between the MDD and control groups were observed only
during the N stimulation (p o0.05), the LZC differences between
the two groups were found in all periods. A topographical map of
average LZC complexity values at each brain site not only shows
these differences between patients and controls but also indicates
the general trends in terms of complexity in each group
(Figs. 5 and 6). The average LZC values in all recorded regions in
the MDD patients also tended to be higher than those in the
control subjects.
To investigate the effectiveness of EEG complexity results for
discriminating MDD patients from healthy controls, ROCs (receiver
operating curves) of each complexity measure were obtained
during each experimental period in each region (Fig. 7). This gure
shows the most distinctive complexity measures as a combination
of period and brain regions. It was found that areas under the ROC
curve (AUC) values of KFD and HFD measures were higher during
emotional stimulation periods in only two brain regions (frontal
and parietal) when compared to resting baseline states.

4. Discussion

Fig. 4. Comparison of HFD values during each experimental period (Abbreviations:


R1 resting 1, M musical stimulus, R2 resting 2, N noise stimulus, R3 resting
3) between control and MDD groups in frontal, central, parietal and temporal
regions. * and ** indicate the statistical differences between groups of p o0.05 and
p o0.001, respectively.

Cognitive disturbances and emotional impairments are common


characteristics of MDD patients [56]. A growing number of studies
have investigated the emotional processing of MDD patients in
vocabulary stimulation or in recognition of emotional facial expressions [38,57-59]. According to these studies, MDD patients showed
impaired emotional processing during positive and negative stimuli.
In another study, Punkanen et al. [35] found that MDD patients
showed a negative bias in emotional evaluations of applied musical
stimuli when compared to control subjects, meaning that they
demonstrated higher scores for unpleasant stimuli (fear, anger, and
sadness) and lower scores for pleasant (happy) stimuli. Although
Punkanen et al. suggested that depression affects the evaluating
process of emotions in patients, they used only subjective rating
scores. The present study investigates different complexity estimators, such as ShEn, LZC, KC, KFD and HFD, in discriminating the MDD
patients from healthy controls both at rest and during emotional
periods. Table 4 presents a brief summary of EEG studies with more
than ten nonlinear features in MDD patients in the literature. To our

Table 2
The p-values (F-values) of estimated complexity features of frontal and central regions of EEG, in discriminating MDD and control groups. (Abbreviations: R1: resting 1,
M: music, R2: resting 2, N: noise, R3: resting 3).
Frontal

R1
M
R2
N
R3

Central

KFD

HFD

ShEn

LZC

KC

KFD

HFD

ShEn

LZC

KC

0.005* (10.498)
0.000* (55.595)
0.000* (52.841)
0.000* (62.291)
0.068 (3.801)

0.026* (5.930)
0.004* (10.720)
0.001* (15.189)
0.000* (18.455)
0.056 (4.216)

0.322 (1.042)
0.217 (1.641)
0.328 (1.015)
0.319 (1.054)
0.799 (0.067)

0.000* (25.495)
0.000* (125.07)
0.006* (9.999)
0.016* (7.121)
0.047* (4.608)

0.273 (1.285)
0.052 (4.371)
0.024* (6.180)
0.024* (6.099)
0.005* (10.266)

0.405 (0.728)
0.341 (0.961)
0.078 (4.321)
0.160 (2.162)
0.152 (2.249)

0.731 (0.122)
0.904 (0.015)
0.119 (2.695)
0.307 (1.109)
0.167 (2.086)

0.369 (0.852)
0.237 (1.500)
0.065 (4.390)
0.146 (2.317)
0.116 (2.742)

0.030* (5.600)
0.138 (2.423)
0.271 (1.295)
0.448 (0.604)
0.448 (0.604)

0.032* (5.454)
0.130 (2.553)
0.265 (1.312)
0.460 (0.518)
0.443 (0.596)

* Less than the signicance p-value of 0.05.

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

55

Table 3
The p-values (F-values) of estimated complexity features of temporal and parietal regions of EEG, in discriminating MDD and control groups (Abbreviations: R1: resting 1, M:
music, R2: resting 2, N: noise, R3: resting 3).
Temporal

R1
M
R2
N
R3

Parietal

KFD

HFD

0.801 (0.066)
0.690 (0.164)
0.588 (0.305)
0.451 (0.596)
0.058 (4.503)

0.898
0.942
0.080
0.320
0.065

ShEn
(0.017)
(0.006)
(3.455)
(1.049)
(4.321)

0.729
0.703
0.887
0.501
0.093

LZC
(0.124)
(0.150)
(0.021)
(0.473)
(3.163)

0.038
0.070
0.070
0.334
0.334

KC
(5.090)
(3.731)
(3.731)
(0.987)
(0.987)

0.032
0.072
0.071
0.332
0.330

(5.120)
(3.725)
(3.730)
(0.989)
(0.992)

KFD

HFD

ShEn

LZC

KC

0.002* (12.677)
0.000* (25.041)
0.033* (5.403)
0.002* (13.115)
0.058 (4.134)

0.042* (4.831)
0.005* (10.404)
0.264 (1.337)
0.805 (0.063)
0.921 (0.010)

0.700 (0.153)
0.572 (0.333)
0.443 (0.617)
0.518 (0.436)
0.403 (0.736)

0.024* (6.103)
0.037* (5.150)
0.098 (3.063)
0.007* (9.373)
0.133 (2.485)

0.030* (5.998)
0.074 (3.611)
0.074 (3.611)
0.074 (3.611)
0.108 (2.887)

* Less than the signicance p-value of 0.05.

knowledge, this is the rst study describing emotion-induced changes in EEG complexity in MDD patients.
The main ndings in the present work were that MDD patients
showed increased EEG complexities (KFD, HFD and LZC) compared
to controls both at rest and during emotional stimulation periods.
This result is consistent with the ndings of past studies, which
reported increased complexities in the brain of patients in a
resting condition [4,28]. According to previous neuro-imaging
studies, this higher complexity has been reported to be related
to the changes in brain dynamics caused by decreased white
matter and increased gray matter in the frontal region of MDD
patients [28,60,61]. Moreover, the increased complexities in MDD
patients may be related to the increased variability, or irregularity, of their EEG data. Second, it has been found that the complexities are lower when only control subjects receive musical
stimulation compared to the resting baseline period. This means
that the EEG signals become less complex/random when the participant is subjected to musical stimulation. This decrease in EEG
dynamic complexities might be explained by strongly coupled
oscillators or the inactivation of previously active neurons and
interacting networks [14,62]. In other words, the displayed values
of KFD, HFD and LZC suggest that listening to CTM reduces the
related region complexities of EEG dynamics in controls in terms
of degrees of randomness or degrees of freedom. A similar conclusion was drawn by a previous study that investigated the effect
of Mozart's music in learning [63]. In that study, less complex EEG
activities were observed in healty controls who were solving
spatial rotation tasks while listening to Mozart's music. These
nding suggest that this type of music enhances the learning of
applied tasks by reducing the brain complexity. Furthermore,
Lamberts et al. [64] and Natarajan et al. [65] showed that subjects
had lower EEG complexities when their brains went to a passive or
relaxed state. Considering these ndings, it may be inferred that
the applied musical stimulation causes a decrease in the complexity level of healthy controls' brains, which indicates that the
subjects are more relaxed during this period. Furthermore, this
result can be explained by the self-rating scores of participants for
evaluating the emotional content of both stimuli after the
experiments. According to results of Pearson's Correlation analysis,
LZC complexities during the music stimulus showed a signicant
negative correlation with the self-rated emotion data of the control subjects (p o0.05). It means that brain complexities of subjects decrease signicantly as the subjects rate the music to be
more positive. On the other hand, we determined that as the
rating scores during noise stimulation decreased, the KFD complexities increased in the frontal cortex. This nding may be
related to the negative potentiation hypothesis, which suggests an
increased emotional reactivity to negative emotional stimuli in
MDD patients. According to the results of an extensive metaanalysis of fMRI studies in MDD patients [66], it is believed that
multiple levels of emotional processing, including the amygdale,
striatum, and prefrontal cortex, are altered in patients' brain. MDD

patients showed generally increased KFD, HFD and LZC complexity


values from baseline into the musical stimulation period, especially in the frontal cortex. In contrast, these measures in control
subjects showed similar values from baseline to the music period.
Therefore, they may show an emotion-specic condition. The
decreased regularities and increased complexities in the EEG signals of MDD patients during music stimulation could be explained
by an increase in the dynamic complexity of neural activity in this
region. On the other hand, this result might be explained by the
observed cognitive impairment of patients, which is associated
with the regulation and processing of emotional and cognitive
information [67]. Several studies have indicated that patients with
MDD display hyperactivity in the brain regions that is critical for
cognitive control (prefrontal and frontal cortices) during cognitive
tasks in comparison with healthy controls [68,69]. Therefore, the
increase in complexity in patients could be due to the increased
neural activation in the brain regions implicated in emotional
processing, and this result may be related to the decits in cognitive performance that were reported by Papazacharias and
Nardini [67]. These differences between patients and controls in
the frontal region are consistent with the ndings of the effects of
listening to music on neural activity, especially in the prefrontal
and frontal cortices [70,71]. Moreover, the HFD complexity values
in the MDD group were not signicantly lower in the M period
compared to the baseline period over the parietal region.
Third, we found an increase in EEG complexities in both groups
when subjects listened to noise compared to musical stimulation. It
has been reported that the complexities of EEGs are mostly associated with the cognitive activity of the brain [72,73]. In our study,
as a cognitive source or task, auditory periods with emotional
content were used. MDD patients exhibited signicantly higher EEG
complexities during the noise period than the music period compared to control subjects. This might be related to the previous
hypothesis suggesting that MDD patients have a negative emotional
bias and exhibit stronger emotional behaviors/ratings when processing negative emotional stimuli [36,74,75]. In our case, we used
the noise period as a disturbing, annoying and negative emotional
stimulus. Therefore, this interpretation could also be accepted as
being in line with the cognitive theory of depression, introduced by
Beck [76], that claried the patients' dysfunctional behaviors during
stressful/negative conditions. A previous study by Gotlib et al. [38]
found that MDD patients paid more attention to negative emotional
faces than positive ones. Therefore, the reasons for this increased
brain complexity during noise might be related to the neurophysiological decits of patients in attention or emotional processing.
Moreover, it is in line with the ndings of a previous study that
reported intensied emotional responses of MDD patients to
negative musical stimulus [77]. Additionally, the increase in EEG
complexity may be explained by the increased cognitive activity in
patients' brains during the noise period. Furthermore, we found
higher degrees of EEG complexities in patients while listening to
noise than in healthy controls. Although MDD patients had

56

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

R1
1

M
1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

R2
1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

R3
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

Fig. 5. Spatial distribution of average LZC values of the EEG in the control subjects during each period.

signicantly different EEG complexities than control subjects during


emotional periods, there was not a statistically signicant difference
in complexity values (except the LZC values of the frontal region)

between the two groups during the resting period (R3). After the
emotional periods, the complexity values generally tended to revert
to the levels observed before the stimulation periods in both

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

57

R1
1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

R2

1
1
0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6
0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

R3
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

Fig. 6. Spatial distribution of average LZC values of the EEG in the MDD patients during each period.

groups. That is, after the noise period, EEG complexities decreased.
Moreover, the EEG complexities particularly decreased in the frontal
(by KFD) and parietal (by HFD) regions when the results of the R1

and R3 periods were compared in MDD patients. In contrast, there


were no signicant complexity differences between the R1 and R3
periods in the control subjects. Overall, we also found that the LZC

D2: correlation dimension; LZC: Lempel-Ziv complexity; KFD: Katz's fractal dimension; HFD: Higuchi's fractal dimension; PNN: probabilistic neural network; RWE: relative wavelet energy; ANN: articial neural network; STARC:
spatiotemporal analysis of relative convergence; DFA: detrended uctuation analysis; L1: Lyapunov exponents; kNN: k-nearest neighbors; LDA: linear discriminant analysis; LR: logistic regression; WPD: wavelet packet
decomposition; WE: wavelet entropy; ApEn: approximate entropy; FD: fractal dimension; ShEn: Shannon entropy; KC: Kolmogorov complexity.

Statistical analysis and PNN Classication accuracy (91.3%) based on HFD values of EEG
ANN
Classication accuracy (98.11%)
Statistical analysis
Signicant differences between male and female MDD patients
KNN, LDA, and LR
Classication accuracy (90%)
Statistical analysis
A good sensitivity for detection of depression based on HFD
PNN
Classication accuracy (99.5%)
Statistical analysis
Controls had higher entropies
Statistical analysis
Signicant differences in LZC between MDD patients and controls
Statistical analysis
Signicant differences between patients and controls and between emotional
periods
KFD and HFD
RWE and entropy
STARC
HFD, D2, DFA, and L1
Spectral asymmetry and HFD
WPD and entropy
WE and ApEn
LZC
KFD, HFD, ShEn, LZC, and KC

A decrease in brain complexity in MDD patients


MDD patients had higher LZC than controls
Statistical analysis
Statistical analysis

Resting EEG
EEG at rest and performing a mental arithmetic
task
Ahmadlou et al. [28]
Resting EEG
Puthankattil and Joseph [30] Resting EEG
Ahmadlou et al. [29]
Resting EEG
Hosseinifard et al. [32]
Resting EEG
Bachmann et al. [34]
Resting EEG
Faust et al. [33]
Resting EEG
Puthankattil and Joseph [31] Resting EEG
Bachmann et al. [4]
Resting EEG
Our study
EEG at rest and during emotional periods

Comparison methods
Extracted features
Data recording
Authors

Table 4
A brief summary of nonlinear EEG studies in MDD patients.

values of the MDD and control groups were closer after the stimulation periods (R3) than at rst baseline (R1). This nding might
be related to increased complexities (LZC and HFD) in control
subjects during the noise stimulus.
When the two groups were compared, it was found that
patients with MDD showed signicantly higher KFD, HFD and LZC
values in the frontal and parietal areas. The importance of the
frontal lobes in the pathophysiology of depression has been well
documented by previous neuro-imaging studies [78-80]. It has
been found that the frontal cortex is dysfunctional in depression,
and patients have reduced prefrontal and frontal lobe volumes
[81]. Studies have also shown that the parietal lobes contribute to
the pathology of MDD [82-85]. Previous studies reported
decreased volumes of the parietal regions in patients [83,85,86]. In
addition, some of these studies found abnormalities in the white
matter structures of the parietal cortex in patients with MDD
[82,84]. Considering all of these ndings, it may be hypothesized
that the changes in brain dynamics found in the nonlinear EEG
analysis result from neurological variances among some cortical
regions of patients with MDD [28]. In a previous thesis study that
investigated the nonlinear dynamics of brain signals in healthy
musician and non-musician participants, higher complexity values
in musicians compared to non-musicians were reported [87]. The
author proposed a possible relationship between these ndings
and anatomical differences between the two groups, which were
observed as increased grey and white matter volumes in musicians
[88,89]. On the other hand, patients and control subjects did not
differ in terms of EEG complexities in the temporal and central
lobes (Tables 2 and 3). For example, Table 2 shows the p-values of
extracted ve EEG features (ShEn, KFD, HFD, LZC, and KC) of the
frontal and central regions for discriminating between the two
groups. As shown in this table, there is no signicant p-value in the
central region. Similarly, none of the obtained p-values are signicant in the temporal region (Table 3). Therefore, it can be
reported that, regardless of the nonlinear method applied, no
signicant complexity differences were found between the EEGs of
MDD patients and controls in the temporal and central regions.
In a previous comprehensive study on schizophrenia patients
[20], the successes of different complexity estimators, such as
ShEn, ApEn, spectral entropy, LZC, and HFD, were compared to
obtain an optimum discrimination between patients and control
subjects. These authors found that the LZC, ApEn, and HFD measures of EEG were more optimal for discrimination. In our study,
the high AUC values of the ROC curve conrmed the ability of the

Results

Fig. 7. ROC curve analysis of KFD and HFD values obtained in parietal and frontal
regions during emotional periods.

D2
LZC

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

Nandrino et al. [27]


Li et al. [19]

58

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

KFD and HFD predictors in the frontal and parietal regions for
distinguishing between patients and controls.

59

Acknowledgments
This study was funded by TBTAK (The Scientic and Technical
Research Council of Turkey) under project number 112E317.

5. Conclusion
In this paper, we have calculated different EEG complexity
measures both at rest and during emotional states in MDD
patients and in control subjects to better understand the underlying brain dynamics. We found that the KFD, HFD and LZC values
were more sensitive in detecting EEG complexities and in discriminating between two groups than the ShEn and KC values,
according to the results of ANOVA. We conclude that there were
differences in the EEG complexities between resting and emotional stimulation periods; that both the groups presented an
increase in the EEG complexities while listening to noise; and that
there was a decrease in brain complexity when control subjects
receive musical stimulation compared to the resting baseline
period. In terms of EEG complexities, better discrimination rate
were obtained using the KFD, HFD and LZC algorithms than the KC
and ShEn methods. Therefore, it can be reported that the ShEn and
KC methods might not be good nonlinear features for the EEG data
in our study. It was found that the emotional processing in the
patients' brains were quantitatively more complicated and complex. The brains of MDD patients had higher complexities than
healthy controls when they were expose to noise stimulus as a
negative and stressful period. MDD patients' negative emotional
bias was shown in the nonlinear EEG analysis.
Although we found that the nonlinear EEG analysis showed
signicant differences between patients with MDD and controls
during both resting and emotional stimulation states, some limitations must be taken into account when planning future studies.
First, the sample size of the participants in our research was small.
To prove the usefulness of these complexity measures in EEG
signals as a diagnostic tool for MDD, these analyses should be
performed on much larger populations. Moreover, further studies
must be carried out to investigate whether the observed complexity variances in the EEG during different emotional periods are
specic to MDD patients. In particular, it will also be important to
compare observed ndings in different pathological or psychological disorders such as bipolar depression, schizophrenia, and
epilepsy. Moreover, the complexity changes produced by emotional periods should be studied by taking into account the
symptom severities, gender effects, and medication status of
patients with MDD.
Future classication studies need to investigate and conrm
these results using different complexity estimators, such as sample
entropy, ApEn, Hurst's exponent, recurrence quantication analysis and higher-order spectra. The experimental results in this study
show the effectiveness of some nonlinear measures for investigating the dynamical behavior of patients' brains during emotional
periods. Because an increasing number of studies have investigated the diagnostic ability of different nonlinear EEG features in
MDD patients, all ndings, including ours, for clarifying the
underlying dynamical activity of the brain may provide objective
criteria in clinical practice.

Conict of interest statement


There is no conict of interest for this study.

Appendix A. Supplementary material


Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.compbiomed.
2015.09.019.

References
[1] R.C. Kessler, E.E. Walters, The national comorbidity survey, in: M.T. Tsaung,
M. Tohen (Eds.), Textbook in Psychiatric Epidemiology, 2nd ed.,John Wiley &
Sons, New York, NY, 2002, pp. 343362.
[2] G. Andrews, Dimensionality and the category of major depressive episode, Int.
J. Methods Psychiatr. Res. 16 (2007) 4151.
[3] American Psychiatric Association (APA), Diagnostic and Statistical Manual of
Mental Disorders, 4th edition, APA, Washington, DC, 1994 DSM-IV.
[4] M. Bachmann, K. Kalev, A. Suhhova, J. Lass, H. Hinkrikus, Lempel Ziv Complexity of EEG in Depression, in: Proceedings of IFMBE, vol 45, 2015, pp. 5861.
[5] S.A. Reid, L.M. Duke, J.J.B. Allen, Resting frontal electroencephalographic
asymmetry in depression: inconsistencies suggest the need to identify mediating factors, Psychophysiology 35 (04) (1998) 389404.
[6] J.J.B. Allen, H.L. Urry, S.K. Hitt, J.A. Coan, The stability of resting frontal electroencephalographic asymmetry in depression, Psychophysiology 41 (2)
(2004) 269280.
[7] A. Sumich, A. Harris, G. Flynn, T. Whitford, N. Tunstall, V. Kumari, et al., Eventrelated potential correlates of depression, insight and negative symptoms in males
with recent-onset psychosis, Clin. Neurophysiol. 117 (8) (2006) 17151727.
[8] M. Bares, M. Brunovsky, M. Kopecek, P. Stopkova, T. Novak, J. Kozeny, et al.,
Changes in QEEG prefrontal cordance as a predictor of response to antidepressants in patients with treatment resistant depressive disorder: a pilot
study, J. Psychiatr. Res. 41 (2007) 319325.
[9] U.R. Acharya, V.K. Sudarshan, H. Adeli, J. Santhosh, J.E.W. Koh, A. Adeli,
Computer-aided diagnosis of depression using EEG signals, Eur. Neurol. 73
(2015) 329336.
[10] C.J. Stam, Nonlinear dynamical analysis of EEG and MEG: review of an emerging eld, Clin. Neurophysiol. 116 (2005) 22662301.
[11] C. Besthorn, R. Zerfass, C. Geiger-Kabisch, H. Sattel, S. Daniel, U. SchreiterGasser, et al., Discrimination of Alzheimer's disease and normal aging by EEG
data, Electroencephalogr. Clin. Neurophysiol. 103 (1997) 241248.
[12] D. Abasolo, R. Hornero, C. Gomez, M. Garcia, M. Lopez, Analysis of EEG background activity in Alzheimers disease patients with Lempel-Ziv complexity
and central tendency measure, Med. Eng. Phys. 28 (2006) 315322.
[13] D. Labate, F. La Foresta, G. Morabito, I. Palamara, F.C. Morabito, Alzheimer's
disease through a multivariate multiscale approach, IEEE Sens. J. 13 (2013)
32843292.
[14] J. Jeong, Nonlinear dynamics of EEG in Alzheimer's disease, Drug Dev. Res. 56
(2) (2002) 5766.
[15] R. Hornero, P. Espino, A. Alonso, M. Lpez, Estimating complexity from EEG
background activity of epileptic patients, IEEE Eng. Med. Biol. 18 (1999) 7379.
[16] S.F. Liang, H.C. Wang, W.L. Chang, Combination of EEG Complexity and Spectral Analysis for Epilepsy Diagnosis and Seizure Detection, EURASIP Journal on
Advances in Signal Processing (2010) 853434.
[17] P. Bob, R. Roman, M. Svetlak, M. Kukleta, J. Chladek, M. Brazdil, Preictal dynamics
of EEG complexity in intracranially recorded epileptic seizure: a case report,
Medicine 93 (23) (2014) 14, http://dx.doi.org/10.1097/MD.0000000000000151.
[18] T. Elbert, W. Lutzenberger, B. Rockstroh, P. Berg, R. Cohen, Physical aspects of
the EEG in schizophrenics, Biol. Psychiatry 32 (7) (1992) 595606.
[19] Y. Li, S. Tong, D. Liu, Y. Gai, X. Wang, J. Wang, et al., Abnormal EEG complexity
in patients with schizophrenia and depression, Clin. Neurophysiol. 119 (2008)
12321241.
[20] M. Sabeti, S. Katebi, R. Boostani, Entropy and complexity measures for EEG
signal classication of schizophrenic and control participants, Artif. Intell.
Med. 47 (2009) 263274.
[21] S. Aydn, N. Arca, E. Ergl, O. Tan, Classication of obsessive compulsive disorder by EEG complexity and hemispheric dependency measurements, Int. J.
Neural Syst. 25 (3) (2015), http://dx.doi.org/10.1142/S0129065715500100.
[22] G. Tononi, G.M. Edelman, O. Sporns, Complexity and coherency: integrating
information in the brain, Trends Cogn. Sci. 2 (1998) 474484.
[23] A. Fernandez, C. Gomez, R. Hornero, J.J. Lopez-Ibor, Complexity and Schizophrenia, Prog. Neuro-psychopharmacol. Biol. Psychiatry 45 (2012) 267276,
http://dx.doi.org/10.1016/j.pnpbp.2012.03.015.
[24] S.A. Akar, S. Kara, F. Latifolu, V. Bilgi, Investigation of the noise effect on
fractal dimension of EEG in schizophrenia patients using wavelet and SSAbased approaches, Biomed. Signal Process. Control 18 (2015) 4248.
[25] Z. Liang, Y. Wang, X. Sun, et al., EEG entropy measures in anesthesia, Front.
Comput. Neurosci. 9 (2015), article 16, 1-17.

60

S. Akdemir Akar et al. / Computers in Biology and Medicine 67 (2015) 4960

[26] R. Esteller, G. Vachtsevanos, J. Echauz, B. Litt, A comparison of waveform


fractal dimension algorithms, IEEE Trans. Circuits Syst. 48 (2001) 177183.
[27] J.L. Nandrino, L. Pezard, J. Martinerie, F. el Massioui, B. Renault, R. Jouvent,
et al., Decrease of complexity in EEG as a symptom of depression, Neuroreport
5 (4) (1994) 528530.
[28] M. Ahmadlou, A. Hojjat, A. Adeli, Fractality analysis of frontal brain in major
depressive disorder, Int. J. Psychophysiol. 85 (2012) 206211.
[29] M. Ahmadlou, H. Adeli, A. Adeli, Spatiotemporal analysis of relative convergence of EEGs reveals differences between brain dynamics of depressive
women and men, Clin. EEG Neurosci. 44 (2013) 175181.
[30] D.S. Puthankattil, P.K. Joseph, Classication of EEG signals in normal and
depression conditions by ANN using RWE and signal entropy, J Mech. Med.
Biol. 12 (2012) 1240019.
[31] S.D. Puthankattil, P.K. Joseph, Analysis of EEG signals using wavelet entropy
and approximate entropy: a case study on depression patients, Int. J. Med.
Health Biomed. Pharm. Eng. 8 (7) (2014) 420424.
[32] B. Hosseinifard, M.H. Moradi, R. Rostami, Classifying depression patients and
normal subjects using machine learning techniques and nonlinear features
from EEG signal, Comput. Methods Programs Biomed. 109 (2013) 339345.
[33] O. Faust, P.A.C. Ang, S.D. Puthankattil, P.K. Joseph, Depression diagnosis support system based on EEG signal entropies, J. Mech. Med. Biol. 14 (2014)
1450035.
[34] M. Bachmann, J. Lass, A. Suhhova, H. Hinrikus, Spectral asymmetry and
Higuchi's fractal dimension measures of depression electroencephalogram,
Comput. Math. Methods Med. 2013 (2013), Article ID 251638, 8 pp.
[35] M. Punkanen, T. Eerola, J. Erkkila, Biased emotional recognition in depression:
perception of emotions in music by depressed patients, J. Affect. Disord. 130
(2011) 118126.
[36] C. Naranjo, C. Kornreich, S. Campanella, et al., Major depression is associated
with impaired processing of emotion in music as well as in facial and vocal
stimuli, J. Affect. Disord. 128 (3) (2011) 243251.
[37] Y. Li, D. Cao, L. Wei, Y. Tang, J. Wang, Abnormal functional connectivity of EEG
gamma band in patients with depression during emotional face processing,
Clin. Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.12.026.
[38] I.H. Gotlib, E. Krasnoperova, D.N. Yue, J. Joormann, Attentional biases for
negative interpersonal stimuli in clinical depression, J. Abnorm. Psychol. 113
(2004) 127.
[39] T. Canli, H. Sivers, M.E. Thomason, S. Whiteld-Gabrieli, J.D. Gabrieli, et al.,
Brain activation to emotional words in depressed vs healthy subjects, Neuroreport 15 (2004) 25852588.
[40] S. Surguladze, M.J. Brammer, P. Keedwell, V. Giampietro, A.W. Young, et al., A
differential pattern of neural response toward sad versus happy facial
expressions in major depressive disorder, Biol. Psychiatry 57 (2005) 201209.
[41] J. Epstein, H. Pan, J.H. Kocsis, Y. Yang, T. Butler, et al., Lack of ventral striatal
response to positive stimuli in depressed versus normal subjects, Am. J. Psychiatry 163 (2006) 17841790.
[42] J. Rottenberg, J.J. Gross, I.H. Gotlib, Emotion context insensitivity in major
depressive disorder, J. Abnorm. Psychol. 114 (4) (2005) 627639.
[43] T. Costa, F. Cauda, M. Crini, M.K. Tatu, A. Celeghin, B. Gelder, M. Tamietto,
Temporal and spatial neural dynamics in the perception of basic emotions
from complex scenes, SCAN 9 (2014) 16901703.
[44] R. Kalender, Trk Musikisinde Kullanilan Makamlarin Tesirleri, Ankara niversitesi Dergisi (1987), p. 361375 , Turkish.
[45] P. Somakci, Trklerde Mzikle Tedavi, Sosyal Bilimler Enstits Dergisi 2
(2003) 131140 , Turkish.
[46] E. Olbrich, The sleeping brain as a dynamical system insights from time
series analysis of the human sleep EEG, in: Proceedings of Biosignal, Berlin,
Germany, 1416 July 2010.
[47] C.E. Shannon, W. Weaver, The Mathematical Theory of Communication., The
University of Illinois Press, Urbana, Illinois, ISBN 0-252-72548-4, 1949.
[48] N. Kannathal, M.L. Choo, U.R. Acharya, P.K. Sadasivan, Entropies for detection
of epilepsy in EEG, Comput. Methods Prog. Biomed. 80 (3) (2005) 187194.
[49] M. Katz, Fractals and the analysis of waveforms, Comput. Biol. Med. 18 (3)
(1998) 145156.
[50] T. Higuchi, Approach to an irregular time series on the basis of the fractal
theory, Phys. D: Nonlinear Phenom. 31 (1988) 277283.
[51] A. Lempel, J. Ziv, On the complexity of nite sequences, IEEE Trans. Inf. Theory
22 (1976) 7581.
[52] X.S. Zhang, R.J. Roy, E.W. Jensen, EEG complexity as a measure of depth of
anesthesia for patients, IEEE Trans. Biomed. Eng. 48 (2001) 14241433.
[53] A.N. Kolmogorov, Logical basis for information theory and probability theory,
IEEE Trans. Inf. Theory 14 (1968) 662664.
[54] G.J. Chaitin, Algorithmic information theory, IBM J. Res. Dev. 21 (4) (1977)
350359.
[55] A. Petrosian, Kolmogorov complexity of nite sequences and recognition of
different preictal EEG patterns, in: Proceedingd of the 8th IEEE Symposium on
Computer-Based Medical Systems, 1995, pp. 212217.
[56] J. Li, C. Xu, X. Cao, Q. Gao, Y. Wang, Y. Wang, J. Peng, K. Zhang, Abnormal
activation of the occipital lobes during emotion picture processing in major
depressive disorder patients, Neural Regen. Res. 8 (18) (2013) 16931701.
[57] J.B. Henriques, R.J. Davidson, Left frontal hypoactivation in depression, J.
Abnorm. Psychol. 100 (1991) 535545.
[58] R. Davidson, D. Lewis, L. Alloy, D. Amaral, G. Bush, J. Cohen, et al., Neural and
behavioral substrates of mood and mood regulation, Biol. Psychiatry 52 (2002)
478502.

[59] J.D. Herrington, W. Heller, A. Mohanty, A.S. Engels, M.T. Banich, A.G. Webb, G.
A. Miller, Localization of asymmetric brain function in emotion and depression, Psychophysiology 47 (3) (2010) 442454.
[60] G.S. Alexopoulos, D.N. Kiosses, S.J. Choi, C.F. Murphy, K.O. Lim, Frontal white
matter microstructure and treatment response of late-life depression: a preliminary study, Am. J. Psychiatry 159 (2002) 19291932.
[61] M.J. Kim, J.P. Hamilton, I.H. Gotlib, Reduced caudate gray matter volume in
women with major depressive disorder, Psychiatry Res.: Neuroimaging 164 (2)
(2008) 114122.
[62] J. Jeong, D.J. Kim, J.H. Chae, S.Y. Kim, H.J. Ko, I.H. Paik, Non-linear analysis ofthe
EEG of schizophrenics with optimal embedding dimension, Med. Eng. Phys. 20
(1998) 669676.
[63] N. Jausovec, K. Jausovec, I. Gerlic, The inuence of Mozart's music on brain
activity in the process of learning, Clin. Neurophysiol. 117 (2006) 27032714.
[64] J. Lamberts, P.L.C. van den Broek, L. Bener, et al., Correlation dimension of the
human electroencephalogram corresponds with cognitive load, Neuropsychobiology 41 (2000) 149153.
[65] K. Natarajan, R. Acharya, F. Alias, et al., Nonlinear analysis of EEG signals at
different mental states, BioMed. Eng. Online 3 (2004) 111.
[66] N.A. Groenewold, E.M. Opmeer, P. Jonge, A. Aleman, S.G. Costafreda, Emotional
valence modulates brain functional abnormalities in depression: evidence
from meta-analysis of fMRI studies, Neurosci. Behav. Rev. 37 (2013) 153163.
[67] A. Papazacharias, M. Nardini, The relationship between depression and cognitive decits, Psychiatr. Danub. 24 (2012) 179182.
[68] P.B. Fitzgerald, et al., An fMRI study of prefrontal brain activation during
multiple tasks in patients with major depressive disorder, Hum. Brain Mapp.
29 (2008) 490501.
[69] S. Schoning, et al., Working-memory fMRI reveals cingulate hyperactivation in
euthymic major depression, Hum. Brain Mapp. 30 (2009) 27462756.
[70] R. Patterson, S. Uppenkamp, I. Johnsrude, T. Grifths, The processing of temporal
pitch and melody information in auditory cortex, Neuron 36 (4) (2002) 767776.
[71] I. Peretz, R. Zatorre, Brain organization for music processing, Annu. Rev. Psychol. 56 (2005) 89114.
[72] J.W. Bruce, Fractal Physiology and Chaos in Medicine, World Scientic Publishing Co. Pte. Ltd, 1990.
[73] N. Kannathal, R. Acharya, F. Alias, T. Tiboleng, S.K. Puthusserypady, Nonlinear
analysis of EEG signals at different mental states, Biomed. Eng. Online 3 (2004) 7.
[74] S.A. Surguladze, A.W. Young, C. Senior, G. Brebion, M.J. Travis, M.L. Phillips,
Recognition accuracy and response bias to happy and sad facial expressions in
patients with major depression, Neuropsychology 18 (2004) 212218, http:
//dx.doi.org/10.1037/0894-4105.18.2.212.
[75] C. Bourke, K. Douglas, R. Porter, Processing of facial emotion expression in
major depression: a review, Aust. NZ J. Psychiatry 44 (2010) 681696, http:
//dx.doi.org/10.3109/00048674.2010.496359.
[76] A. Beck, The evolution of the cognitive model of depression and its neurobiological correlates, Am. J. Psychiatry 165 (2008) 969977, http://dx.doi.org/
10.1176/appi.ajp.2008.08050721.
[77] E. Bodner, I. Iancu, A. Gilboa, A. Sarel, A. Mazor, D. Amir, Finding words for
emotions: the reactions of patients with major depressive disorder towards
various musical excerpts, Arts Psychother. 34 (2007) 142150.
[78] A.L. Brody, M.W. Barsom, R.G. Bota, S. Saxena, Prefrontalsubcortical and
limbic circuit mediation of major depressive disorder, Semin. Clin. Neuropsychiatry 6 (2001) 102112.
[79] R.J. Davidson, J.B. Henriques, Regional brain function in sadness and depression, in: J. Borod (Ed.), The Neuropsychology of Emotion, Oxford University
Press, New York, NY, 2000, pp. 269297.
[80] D.A. Pizzagalli, J.B. Nitschke, T.R. Oakes, A.M. Hendrick, K.A. Horras, C.L. Larson,
et al., Brain electrical tomography in depression: the importance of symptom
severity, anxiety, and melancholic features, Biol. Psychiatry 52 (2002) 7385,
http://dx.doi.org/10.1016/S0006-3223(02)01313-01316.
[81] S. Bell-McGinty, M.A. Butters, C.C. Meltzer, et al., Brain morphometric
abnormalities in geriatric depression: long-term neurobiological effects of
illness duration, Am. J. Psychiatry 159 (2002) 14241427.
[82] F. Biver, S. Goldman, V. Delvenne, A. Luxen, V. De Maertelaer, P. Hubain,
J. Mendlewicz, F. Lotstra, Frontal and parietal metabolic disturbances in unipolar depression, Biol. Psychiatry 36 (1994) 381388.
[83] M. Ballmaier, A. Kumar, P.M. Thompson, K.L. Narr, H. Lavretsky, L. Estanol,
et al., Localizing gray matter decits in late-onset depression using computational cortical pattern matching methods, Am. J. Psychiatry 161 (2004)
20912099.
[84] N. Ma, L. Li, N. Shu, J. Liu, G. Gong, Z. He, Z. Li, L. Tan, W.S. Stone, Z. Zhang, L. Xu,
T. Jiang, White matter abnormalities in rst-episode, treatment-naive young
adults with major depressive disorder, Am. J. Psychiatry 164 (2007) 823826.
[85] M.A. Ikram, H.J. Luijendijk, M.W. Vernooij, A. Hofman, W.J. Niessen, A. van der
Lugt, et al., Vascular brain disease and depression in the elderly, Epidemiology.
21 (2010) 7881.
[86] W.C. Drevets, J.L. Price, J.R. Simpson, R.D. Todd, T. Reich, M. Vannier, M.
E. Raichle, Subgeual prefrontal cortex abnormalities in mood disorders, Nature
386 (6627) (1997) 827-827.
[87] S.M. Carpentier, Brain-Music Duet: MEG Signal Complexity and Auditory
Perception in Musicians and Nonmusicians, Department of Psychology, University of Toronto, 2011.
[88] G. Schlaug, The brain of musicians, Ann. NY Acad. Sci. 930 (2001) 281299.
[89] C. Gaser, G. Schlaug, Brain structures differ between musicians and nonmusicians, J. Neurosci. 23 (27) (2003) 92409245.