Clinical Neurophysiology 118 (2007) 24892496

www.elsevier.com/locate/clinph

Detrended uctuation analysis of resting EEG in depressed

outpatients and healthy controls
Jun-Seok Lee a, Byung-Hwan Yang b, Jang-Han Lee c,*, Jun-Ho Choi b,
Ihn-Geun Choi d, Sae-Byul Kim e
a

Department of Psychiatry, Kwandong University College of Medicine, Myongji Hospital, 697-24 Hwajeong, Dukyang, Gyang,
Gyunggi 412-270, Republic of Korea
Department of Neuropsychiatry, Hanyang University College of Medicine, 17 Haengdang, Seongdong, Seoul 133-791, Republic of Korea
c
Department of Psychology, Chung-Ang University, 221, Heukseok-dong, Dongjak-gu, Seoul 156-756, Republic of Korea
d
Department of Neuropsychiatry, Hallym University Han-Gang Sacred Heart Hospital, Seoul 150-719, Republic of Korea
e
Korean Intellectual Property Oce, 920 Dunsandong, Seogu, Daejeon 302-701, Republic of Korea
Accepted 4 August 2007
Available online 24 September 2007

Abstract
Objective: Recent ndings have demonstrated that the EEG possesses long-range temporal (auto-) correlations (LRTC) in the dynamics
of broad band oscillations. The analysis of LRTC provides a quantitative index of statistical dependencies in oscillations on dierent time
scales. We analyzed LRTC in resting EEG signals in depressed outpatients and healthy controls.
Methods: The participants in this study were 11 non-depressed, age-matched controls, and 11 unmedicated unipolar depressed patients.
EEG data were obtained from each participant during 5-min resting baseline periods with eyes closed and then analyzed with detrended
uctuation analysis (DFA), a scaling analysis method that quanties a simple parameter to represent the correlation properties of a time
series. The scaling exponent, the result of DFA, provides a quantitative measure of LRTC from the EEG.
Results: The present study demonstrates that all the scaling exponents in depressed patients and healthy controls were greater than 0.5
and less than 1.0, regardless of condition. Furthermore, the scaling exponents of depressed patients have relatively higher values in whole
brain regions compared to healthy controls, with signicant dierences at F3, C3, T3, T4 and O1 channels (p < 0.05). Finally, a significant linear correlation was observed between the severity of depression and the scaling exponent over most of the channels, except O2.
Conclusions: These results suggest that the brain aected by a major depressive disorder shows slower decay of the LRTC, and that the
persistence of the LRTC of EEG in depressed patients was associated with the severity of depression over most of the cortical areas.
Signicance: The DFA method may broaden our understanding of the psychophysiological basis of depression.
 2007 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: Major depressive disorder; EEG; Detrended uctuation analysis

1. Introduction
There is some evidence that the resting EEG may be a
useful tool in the study of depression. Patients diagnosed
with clinical depression have frequently exhibited increased
alpha and/or theta power (Nystrom et al., 1986; Pollock

Corresponding author. Tel.: +82 2 820 5851; fax: +82 2 816 5124.
E-mail address: clipsy@cau.ac.kr (J.-H. Lee).

and Schneider, 1990) as well as interhemispheric asymmetry (Schaer et al., 1983; Henriques and Dividson, 1990)
and hypocoherence in anterior regions (John et al., 1988;
Ford et al., 1986). These ndings for resting EEG in
depressed patients are based on morphologic analysis
which decomposes the component frequencies and amplitudes in the EEG oscillation.
In contrast to these conventional approaches in terms of
linear dynamics, alternative approaches were applied to the
EEG oscillations in the framework of non-linear dynamics.

1388-2457/$32.00  2007 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.clinph.2007.08.001

2490

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

It was assumed that the EEG within a particular psychophysiological state could be described by non-linear
dynamics and, therefore, could be characterized by non-linear analysis. With regard to this perspective, there has been
a sustained interest in analyzing the EEG within the context of non-linear dynamics (Rapp et al., 1989; Rapp,
1993; Palus, 1996). Recent ndings have shown that the
EEG possesses long-range temporal (auto-) correlations
(LRTC) in the dynamics of broad band oscillations (Parish
et al., 2004; Watters and Martin, 2004) and narrow bandpass ltered oscillations (Linkenkaer-Hansen et al., 2001;
Nikulin and Brismar, 2005). The analysis of LRTC provides a quantitative index of statistical dependencies in
oscillations on dierent time scales (as opposed to crosscorrelation analysis, which provides an index of statistical
dependencies between dierent channels) (LinkenkaerHansen et al., 2005). An essential feature of LRTC is their
power-law behavior, which indicates that the mechanisms
contributing to their build-up are similar at dierent time
scales (Nikulin and Brismar, 2005). The importance of
LRTC stems from the fact that their presence is thought
to be advantageous for a reliable transfer of information
in neuronal populations (Chialvo and Bak, 1999; Beggs
and Plenz, 2003).
The characteristics of non-linear dynamics in EEG oscillations in patients with depressive disorder have been noted
in several clinical studies (Nandrino et al., 1994; Thomasson et al., 2000). These studies suggest that mood be considered to be a psychophysiological state emerging from a
specic neuronal process revealed by non-linear analysis
methods. However, there has been a diculty in establishing the nature of the underlying non-linear process more
precisely as it could be due to a number of interrelated
methodological drawbacks. One of these drawbacks is the
dierentiation between non-linear dynamics and noisy
background in a spatially extended system (Grassberger,
1989; Mayer-Kress and Kaneko, 1989; Palus, 1996). In this
context the detrended uctuation analysis (DFA), invented
by Peng et al. (1992), has been established as an important
non-linear analysis technique for the detection of LRTC in
a non-stationary time series. DFAs advantages are its
enabling the detection of LRTC embedded in a non-stationary time series, as well as its avoidance of the spurious
detection of apparent LRTC that are the noise of a nonstationary time series (Kantelhardt et al., 2001; Watters
and Martin, 2004). Several studies have applied DFA to
EEG analysis and have convincingly demonstrated that
EEG has LRTC (Pereda et al., 1998; Popivanov and Mineva, 1999; Linkenkaer-Hansen et al., 2001). Recently, we
have shown that LRTC in EEG by DFA during a hypnotic
state diers signicantly from that with no hypnotic state
(Lee et al., 2007).
Assuming that the depression is related to a particular
psychophysiological state, we analyzed LRTC in resting
EEG signals in depressed outpatients and healthy controls,
using the DFA method. Another important aspect related
to the description of psychophysiological state of depres-

sion is a possible dependency of LRTC on topography

and severity of depression.
2. Methods and materials
2.1. Participants
Data were collected between June 2005 and November
2005 in Myongji Hospital Depression Clinic in Korea. Eleven right-handed unmedicated and unipolar depressed outpatients, 44 11 years of age (nine of which were females),
participated in the study. Illness severity was assessed prior
to EEG data collection. Depression symptoms were assessed
by the Beck depression inventory (BDI; Beck et al., 1961)
and the 17-item Hamilton depression rating scale (HDRS;
Hemilton, 1960). To be classied as depressed, two criteria
were to be met: (1) a BDI score of P10 and a HDRS score
of P14 at intake; and (2) a DSM-IV (American Psychiatric
Association, 1994) interview resulting in a diagnosis of a
major depressive episode.
Eleven right-handed healthy control participants
matched for age and sex were recruited through Internet
advertisement and screened for psychiatric or neurological
disease. To be classied as a non-depressed control, participants needed to score 69 on the BDI. In fact, the range of the
depressed group was 1443, with a mean of 28.36 on the BDI,
while the range of scores among the controls was 38, with a
mean of 4.36. Exclusion criteria for both groups consisted of:
(1) diagnosis of schizophrenia (or other psychotic disorders),
psychopathic personality disorder or alcohol/drug abuse or
dependence; (2) evidence of dementia, delusions, hallucinations or indication of a serious suicidal risk; (3) history of seizures, brain surgery, organic brain disease or organic
aective disturbance; (4) clinically signicant neurologic,
gastrointestinal, hepatic, renal, hematologic or respiratory
disease; (5) electroconvulsive therapy (ECT) history; (6)
use within the past 4 weeks of antidepressants (8 weeks in
the case of uoxetine), benzodiazepines, barbiturates, beta
blockers, cimetidine, clonidine, anticoagulants or any centrally active agent. The study was approved by the Institutional Review Boards of Kwandong University Myongji
Hospital. All participants gave their written informed consent before enrolling in the study.
2.2. EEG recordings and experimental conditions
The participants were seated in a comfortable armchair
in a sound-attenuated, dimly lit EEG room. After having
become accustomed to the atmosphere, the participants
were checked for a 5-min resting EEG collection with eyes
closed. To avoid any disturbances and interferences, the
recording room was shielded with copper. EEG were
recorded using silversilver chloride cup electrodes (Ag
AgCl) attached by collodion, on F3 and F4 (frontal), on
C3 and C4 (central), on T3 and T4 (temporal) and on O1
and O2 (occipital) scalp regions of the 1020 international
system (American Electroencephalographic Society, 1994).

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

All electrode impedances were below 5 kX. The EEG-measurement-device (LEX3208, LAXTHA Inc., Korea) settings were as follows: 256-Hz sampling rate, 12-bit
analog-to-digital (A/D) converter, 0.6-Hz high-pass lter,
46-Hz low-pass lter, and 60-Hz notch lter. The EEG
data were visually inspected for artifacts. Data contaminated with artifact were discarded.
2.3. Data analysis and statistics
The EEG data were successively segmented into consecutive 30 s epochs for DFA. The sequences of 10 30-s epochs of
each EEG served as the database. For each 30 s epoch, we
applied DFA according to the following algorithm (Peng
et al., 1995; Goldberger et al., 1999) and calculated the median value from the results of DFA at each channel. DFA is a
method for quantifying the correlation property in a nonstationary time series by computing a scaling exponent using
a modied root mean square analysis of a random walk.
Briey, the analysis is applied to a discrete time series
x(i), i = {1, . . ., N}, which in the present study represents
the synchronization likelihood averaged over all pairs of
channels for each time sample (N = 7680/30 = 256). In
the rst step, the mean is subtracted from this time series
and the time series is integrated

2491

k
X
yk
xi  hxi
i1

where x is the average of the synchronization likelihood

time series. Next, the de-meaned, integrated time series
y(k) is divided into a number of segments with length n
(n represents the time scale of observation). In this study
we used the following time segment lengths: 0.1, 0.15,
0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55 and 0.6 s (equivalently 25, 37, 49, 61, 73, 85, 97, 109, 121, 133 and 145 samples in segment lengths. The above procedure is illustrated
in Fig. 1. The samples are the number of points, which
should be a positive number for the nature of the analysis.
Therefore, the numbers were rounded o). For each of
these segments, the local least-squares linear t is determined. The ensuing piecewise linear t is designated
yn(k). Then, the integrated time series y(k) is detrended
by subtracting the local linear t yn(k) for each segment.
The root mean square uctuation of this integrated and
detrended time series is given by:
r
1 XN
2
F n
yk  y n k
k1
N

Fig. 1. Time course of instantaneous amplitude of broad band oscillations. The amplitude at 15 Hz is shown for a typical F3 channel in the depressed
patient and healthy control during 30 s (a and b, respectively). Enlarged segment of time interval 03 s, for the depressed patient and healthy control (c and
d, respectively). For Fig. 2 data are presented for the same subject, F3 channel.

2492

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

Subsequently, this determination of F(n) is repeated for

a range of dierent scales n (in the present study n ranged
from 25 samples to 145 samples). In the nal step, the logarithm of F(n) is plotted as a function of the logarithm of
the time scale n (we used logarithms with a base of the
number e). If the time series x(i) has self-similar, scale-free
(fractal) properties, this plot will display a linear scaling
region with a certain scaling exponent. The slope of the line
relating Ln(F(n))/Ln(n) determines the scaling exponent
(self-similarity parameter), which is the result of DFA
(Peng et al., 1995; Goldberger et al., 1999). Fig. 2 shows
an example of DFA for the amplitude envelope of the
broad band oscillations, demonstrating a linear relationship between Ln(n) and Ln(F(n)) in the 030 s range.
The scaling exponent provides a quantitative measure of
LRTC that exist in the EEG. When the EEG is completely
uncorrelated (Gaussian or non-Gaussian probability distribution), the calculation of the scaling exponent yields 0.5,
also called white noise. When applied to the EEG with
LRTC, power-law behavior will generate scaling exponents
with either greater than 0 and less than 0.5 or greater than
0.5 and less than 1. A scaling exponent greater than 0.5 and
less than 1 indicates the data are correlated such that large
uctuations are likely to be followed by large uctuations
and small uctuations are likely to be followed by small
uctuations. The case of scaling exponent = 1 is a special
one which corresponds to 1/f noise, the LRTC became
independent of the time innite range (Bak et al., 1987).
As the scaling exponent increases from 0.5 toward 1, the
LRTC in the EEG are more persistent (decaying more
slowly with time). If a scaling exponent is greater than 1,
the LRTC no longer exhibits power law behavior. Finally,
if the scaling exponent = 1.5, this indicates Brownian noise,
which is the integration of white noise (Peng et al., 2000;
Parish et al., 2004).

Comparisons between depressed patients and healthy

controls were carried out using a t-test for the mean values of the scaling exponents of these two conditions. In
addition, a mixed-model repeated-measures ANOVA (2
within and 1 between) was used to determine signicance. The within-group variables are the recording site
(SITE: measured at F3, F4, C3, C4, T3, T4, O1 and
O2) and recording hemisphere (HEMISPHERE: right
and left hemisphere), and the between-group variable is
the participant group (GROUP: depressed patients and
healthy controls). The Greenhouse-Geisser correction
for multiple factors was adopted to control for the
assumption of sphericity. Finally, correlations between
the severity of depression (BDI scores) and the scaling
exponents were determined with Pearson correlation
coecients. All statistical procedures were performed
using SPSS Version 10.0 for Windows. The signicance
level was set at p < 0.05.
In order to test non-linearity in the original EEG signals, we compared the scaling exponents of the original
data with those of the corresponding surrogate data
(Theiler and Rapp, 1996). The surrogate data were generated herein by rst computing the Fourier transforms
of the original signals, randomizing the Fourier phase
while preserving the moduli, and then performing inverse
Fourier transforms (Ivanov et al., 1996). Surrogate data
have amplitude spectra identical to the original signals,
but possible temporal correlations are destroyed. Eleven
surrogate data were generated to match each original
data of depressive patients. The scaling exponents of
DFA are computed for both original data and surrogate
data. The statistical signicance (S) for testing non-linearity is dened by
S jDFA exponentsurr  DFA exponentorig j=rDFA exponentsurr

DFA exponentsurr denotes the mean value of scaling

exponent of the surrogate data and r (DFA exponentsurr) its standard deviation, while in the following
the scaling exponent of the original data is labeled
DFA exponentorig. This statistic represents the number
of standard deviations (r) of distance from the DFA
exponent of the original data. It follows a Students t
test distribution with 10 degrees of freedom (tg[1a/2]).
For a = 0.05, critical value of t is 2.228. If the value
of S was larger than 2.228, the original data were considered non-linear.
3. Results
3.1. Dierence between the scaling exponents for depressed
patients and healthy controls
Fig. 2. Detrended uctuation analysis (DFA) quanties correlations in
broad band EEG signals. The scaling exponents are 0.918 and 0.614 for
the depressed patient and healthy control, respectively. Same electrode
location and subject as in Fig. 1.

Table 1 provides the mean and the standard deviation of

the scaling exponent at each channel. All of the scaling
exponents in the depressed patients and healthy controls
were greater than 0.5 and less than or equal to 1.0, indicat-

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

Table 1
The scaling exponents of detrended uctuation analysis of resting EEG in
depressed patients (N = 11) and healthy controls (N = 11)
Channel

F3
F4
C3
C4
T3
T4
O1
O2
*

Depressed patient

Healthy control

Mean

SD

Mean

SD

0.96
0.96
0.87
0.86
0.93
0.98
0.82
0.83

0.16
0.18
0.16
0.16
0.14
0.17
0.17
0.20

0.83
0.86
0.75
0.75
0.80
0.84
0.69
0.73

0.11
0.18
0.10
0.09
0.11
0.11
0.10
0.14

t
2.25*
1.29
2.22*
1.95
2.39*
2.36*
2.19*
1.34

Table 2
The scaling exponents of detrended uctuation analysis of resting EEG in
depressed patients (N = 11) and surrogate data (N = 11)
Channel

F3
F4
C3
C4
T3
T4
O1
O2
**

Table 3
Repeated measures ANOVA table for HEMISPHERE main eect
Channel
a

Right hemisphere
Left hemisphereb

Source

SS

DF

MS

ERROR
SITE

0.64
0.31

60
3

.01
.10

9.75

0.001*

ERROR
SITE

0.19
0.24

50.30
2.52

.004
.095

24.97

0.001*

Right hemisphere: sphericity assumed.

Left hemisphere: sphericity violated, Greenhouse-Geisser corrections
to the D.F. were used to adjust for violations of the sphericity assumption
for repeated measures factors.
*
p < 0.01.
b

p < 0.05.

ing persistent LRTC of the original data sequence. The

scaling exponents of depressed patients have relatively
higher values in whole brain regions compared to healthy
controls, with signicant dierences at F3, C3, T3, T4
and O1 channels.
The mean and the standard deviation of the scaling
exponent for depressed patients and surrogate data is
also given in Table 2. The scaling exponents of surrogate
data have signicantly higher values compared with
depressed patients at all channels. The demonstration
of a statistically signicant dierence in scaling exponents
between the original and surrogate data is in keeping
with the presence of non-linear dynamics in the original
signal.
In order to answer the question of whether there is a
global dierence in LRTC between depressed patients
and healthy controls or not, repeated measures ANOVA
was performed for two groups (GROUP: depressed
patients and healthy controls) eight channels (SITE:
measured at F3, F4, C3, C4, T3, T4, O1 and O2). In
the analysis, with F values representing the main eect
of the group (F(1, 20) = 5.15, p < 0.05) and site
(F(7, 140) = 12.33, p < 0.01), the global null hypothesis
that there is no dierence between depressed patients
and healthy controls was rejected. Nevertheless, there
was no signicant interaction of group and site
(F(7, 140) = 0.25, p > 0.05). This suggests that the global

2493

p < 0.05.
p < 0.01.

Depressed patient

Surrogate data

Mean

SD

Mean

SD

0.96
0.96
0.87
0.86
0.93
0.98
0.82
0.83

0.16
0.18
0.16
0.16
0.14
0.17
0.17
0.20

1.08
1.06
1.01
0.99
1.06
1.13
0.93
0.92

0.25
0.23
0.25
0.24
0.19
0.21
0.23
0.21

dierence in LRTC between depressed patients and

healthy controls reects the neuronal dynamics related
to generation of depression rather than an incidental
result of the interaction of the group and site.
Next, in order to test inter-hemispheric dierences,
repeated measures ANOVA of LRTC was performed for
two groups (patients and controls) two hemispheres
(F3, C3, T3 and O1 classied as left hemisphere and F4,
C4, T4 and O2 as right). The right hemisphere satised
the sphericity assumption but the left hemisphere violated
this assumption. Thus, a Greenhouse-Geisser correction
was performed in the left hemisphere. A signicant main
eect was observed for both hemispheres (Table 3). Additionally, repeated measures ANOVA was performed for
two groups (patients and controls), considering four channels at each hemisphere. The ANOVA of LRTC yielded a
signicant main eect of the group in the left hemisphere,
F(1, 20) = 5.87, p = 0.025, but a marginal trend in the right
hemisphere, F(1, 20) = 4.14, p = 0.055 (Table 4). This suggests a better discrimination between depressed patients
and healthy controls for the left hemisphere than for the
right.
3.2. Correlations between the BDI and scaling exponents
Finally, to determine the possible eect of severity of
depression on the scaling exponents, the correlations
between the scaling exponents and the BDI score of the
participants were examined. Fig. 3 shows the correlations
between the scaling exponents and the BDI score of the
participants. The BDI score was signicantly correlated
with the scaling exponents at F3, F4, C3, C4, T3, T4 and
O1.

t
3.08*
2.83*
3.83*
2.76*
3.79**
3.79**
2.26*
2.97*

Table 4
Repeated measures ANOVA table for GROUP main eect
Channel

Source

SS

DF

MS

F3, C3, T3 and O1

ERROR
GROUP

1.26
0.37

20
1

0.06
0.37

5.87

0.025*

ERROR
GROUP

1.32
0.27

20
1

0.07
0.27

4.14

0.055

F4, C4, T4 and O2

*

p < 0.05.

2494

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

Fig. 3. Correlations between the DFA power-law scaling exponents of resting EEG and severity of depression of total participants (N = 22). BDI, Beck
depression inventory; *p < 0.05, **p < 0.01.

4. Discussion
In this study we analyzed the resting EEG of 11 unmedicated depressed patients and 11 healthy control participants by means of DFA, a scaling analysis method that
quanties a simple parameter to represent the correlation
properties of a time series. The scaling exponent, the result
of DFA, provides a quantitative measure of LRTC that
exists in the EEG. The present study demonstrates that
all the scaling exponents in depressed patients and healthy
controls were greater than 0.5 and less than 1.0, regardless
of their condition. These results show that the broadband
EEG in both depressed patients and healthy controls exhibit LRTC with power-law behavior.
Several studies have applied DFA to broadband EEG
and have found indications of LRTC with power-law
behavior. Watters and Martin (2004) reported that a mean
scaling exponent of 0.67 was the result of DFA applied to
the EEG obtained from 3 to 100 Hz, and our previous
study (Lee et al., 2007) reported a score of 0.91 from psychiatric outpatients. The main reasons for this dierence
are presumed to be that all of the participants of the latter
study had psychopathologies and were not controlled by
psychiatric medications. In this study, the mean scaling
exponent across all participants and sites of EEG was
0.84, which is located between the result of Watters and
Martin (2004) and the result obtained by our previous
study.
For the EEG data, non-linear dynamics algorithms can
give spurious results. Recently, the surrogate data techniques have been developed to distinguish the non-linear
systems from the linearly correlated noise (Theiler and
Rapp, 1996). We found that the scaling exponents of surrogate data have signicantly higher values compared with
depressed patients at all channels. That is a strong indication of the non-linear structure in the original EEG data.
From the above, it is apparent that LRTC with powerlaw scaling is a fundamental feature of brain activity. Further, the presence of power-law scaling behavior in the

LRTC of the EEG demonstrates that there are persistent

temporal correlations with large (small) uctuations that
tend to cluster together in time, as seen in other complex
systems exhibiting power-law scaling behavior (Peng
et al., 1995).
Furthermore, we found that depressed patients demonstrate higher values of the scaling exponents at all channels
than healthy control participants, with signicant dierences detected at F3, C3, T3, T4 and O1 channels. This
result suggests that the EEG oscillations exhibit more persistent tendency of the LRTC in depressive patients compared to healthy controls. Although this dierence does
not implicate a specic underlying mechanism of depression, it suggests that distinct mechanisms underlie the
EEG oscillations. It could be that the mechanism responsible for higher values of the scaling exponents in depressed
patients than healthy controls is responsible for the psychophysiological changes seen in depression.
In this study, the mean scaling exponent of depressed
patients was 0.90 and the value of healthy controls was
0.78. In a previous study that analyzed broadband EEG,
Parish et al. (2004) reported that the scaling exponent of
the epileptogenic hippocampus was 0.680.88 as a result
of DFA applied to intracranial EEG data of the hippocampal area for 5 patients with a unilateral mesial temporal
lobe seizure. On the other hand, it was between 0.60 and
0.69 in the non-epileptogenic hippocampus. In the study
of DFA applied to sleep apnea patients, Lee et al. (2004)
also reported that the scaling exponent increased as sleep
became deeper in that it was 0.95, 1.00, 1.12, 1.22 in waking, stage 1, stage 2, and stage 34 sleep, respectively.
Recently Gifani et al. (2007) applied DFA to EEG to measure the depth of anesthesia and reported that the scaling
exponent was increased from 0.8 to 2 as it changed from
awake to moderate and then deep anesthesia. Connected
to this point, a signicant dierence was found only in
the upper alpha band (1013 Hz) as a result of DFA in Alzheimer disease (AD) patients and non-demented participants with subjective memory complaints (SC). The

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

scaling exponent of the AD group was 0.842, which was

greater than 0.786 of the SC group (Stam et al., 2005).
The pathophysiological implications of slower decay of
the LRTC in the EEG in depression, dementia, and deeper
anesthesia/sleep are not clear. However, major depression,
Alzheimer dementia, anesthesia and sleep are commonly
accompanied by the decline of cognitive function. On the
other hand, cognitive abilities are increasingly linked to
the dynamics of neuronal oscillations on many spatial
scales and at various frequencies (Varela et al., 2001; Buzsaki and Draguhn, 2004). These ndings suggest that the
persistence of LRTC in the EEG is closely related to cognitive ability. Thus, one can speculate that in the case of
depression, slow decay is related to the reduction of cognitive abilities in depressive disorder, such as ruminations
and psychomotor retardation (Widlocher, 1983).
As previously mentioned, the scaling exponent of
depressed patients was signicantly higher than healthy
controls at F3, C3, T3, T4 and O1 channels. Furthermore,
there was no systemic variation across recording channels
for these dierent groups shown only in the left hemisphere. Therefore, our results indicated that depressed
patients have signicantly higher values of the scaling
exponents over the left hemisphere and both temporal
regions than healthy controls.
In 1996, Sheline et al. demonstrated a signicant inverse
correlation between depression and left hippocampal volume. Vakili et al. (2000) also found that left hippocampal
volumes correlated with the HDRS. Using PET, subsequent studies showed that the amount of depletion of serotonin (S2) receptors in the left temporal cortex was
signicantly correlated with the severity of depressive
symptoms (Mayberg et al., 1988). On the other hand, some
investigators have suggested that a left hemispheric epileptic focus may be a factor predisposing to depression (Septien et al., 1993; Altshuler et al., 1990). Thus, the results of
our study are in line with previous studies showing depressive episodes have been associated with temporal structures
(Sheline et al., 1996; Vakili et al., 2000; Mayberg et al.,
1988) and left hemispheric structures (Septien et al., 1993;
Altshuler et al., 1990).
To gain further insights into the basis for the LRTC in
the EEG and major depressive disorder, we tested whether
the scaling exponent was related to the severity of depression in the participants. A signicant linear correlation
was observed between the score in the BDI and the scaling
exponent over most of the channels, except O2. In other
words, a brain aected by major depressive disorder shows
slower decay of the LRTC and the persistence of the LRTC
of EEG in depressed patients was associated with the severity of depression over most of the cortical areas.
This study suggests that further clarication would show
how the DFA method can broaden our understanding of
the psychophysiological basis of depression. Future studies
should focus on researching of the slower decay of LRTC
of EEG changes with the recovery from a depressive episode. Future works should also include the comparison

2495

of the DFA method in the EEG of depressed patients with

other neuroimaging methods.
Acknowledgement
This work was supported by the Korea Research Foundation Grant funded by the Korean Government (KRF2006-E00071).
References
Altshuler LL, Devinsky O, Post RM, Theodore W. Depression, anxiety,
and temporal lobe epilepsy. Arch Neurol 1990;147:2848.
American Electroencephalographic Society. Guideline thirteen: guidelines
for standard electrode position nomenclature. J Clin Neurophysiol
1994;11:1113.
American Psychiatric Association. Diagnostic and statistical manual of
mental disorders, 4th ed. Washington, DC: APA; 1994.
Bak P, Tang C, Wiesenfeld K. Self-organized criticality: an explanation of
the 1/f noise. Phys Rev Lett 1987;59:3814.
Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An inventory for
measuring depression. Arch Gen Psychiatry 1961;4:56171.
Beggs JM, Plenz D. Neuronal avalanches in neocortical circuits. J Eurosci
2003;23:1116777.
Buzsaki G, Draguhn A. Neuronal oscillations in cortical networks.
Science 2004;304:19269.
Chialvo DR, Bak P. Learning from mistakes. Neuroscience
1999;90:113748.
Ford MR, Goethe JW, Dekker DK. EEG coherence and power in the
discrimination of psychiatric disorders and medication eects. Biol
Psychiatry 1986;21:117588.
Gifani P, Rabiee HR, Hashemi MH, Taslimi P, Ghanbari M. Optimal
fractal-scaling analysis of human EEG dynamic for depth of anesthesia
quantication. J Franklin Inst 2007;344:21229.
Goldberger AL, Amaral LAN, Glass L, Hausdor JM, Ivanov P, Mark
RG, et al. PhysioBank, PhysioToolkit, and Physionet: components of
a new research resource for complex physiologic signals. Circulation
1999;101:e21520.
Grassberger P. Information content and predictability of lumped and
distributed dynamical systems. Phys Scr 1989;40:34653.
Hemilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry
1960;23:5662.
Henriques JB, Dividson RJ. Regional brain electrical asymmetries
discriminate between previously depressed and healthy control subjects. J Abnorm Psychol 1990;99:2231.
Ivanov PC, Rosenblum MG, Peng C-K, Mietus J, Havlin S, Stanley HE,
et al. Scanning behaviour of heartbeat intervals obtained by waveletbased time-series analysis. Nature 1996;383:3237.
John ER, Prichep LS, Fridman J. Neurometrics: computer assisted
dierential diagnosis of brain dysfunctions. Science 1988;293:1629.
Kantelhardt J, Koscielny-Bunde E, Rego H, Havlin S, Bunde A. Detecting
long-range correlations with detrended uctuation analysis. Phys A
2001;295:44154.
Lee JM, Kim DJ, Kim IY, Park KS, Kim SI. Nonlinear-analysis of human
sleep EEG using detrended uctuation analysis. Med Eng Phys
2004;26:7736.
Lee JS, Spiegel D, Kim SB, Lee JH, Kim SI, Yang BH, et al. Fractal
analysis of EEG in hypnosis and its relationship with hypnotizability.
Int J Clin Exp Hypn 2007;55:1431.
Linkenkaer-Hansen K, Nikouline VV, Palva JM, Ilmoniemi RJ. Longrange temporal correlations and scaling behavior in human brain
oscillations. J Neurosci 2001;21:13707.
Linkenkaer-Hansen K, Monto S, Rytsala H, Suominen K, Isometsa E,
Kahk}
onen S. Breakdown of long-range temporal correlations in theta
oscillations in patients with major depressive disorder. J Neurosci
2005;25:101317.

2496

J.-S. Lee et al. / Clinical Neurophysiology 118 (2007) 24892496

Mayberg HS, Robinson RG, Wong DF, Parikh R, Bolduc P, Starkstein

SE, et al. PET imaging of cortical S2-serotonin receptors after stroke:
lateralized changes and relationship to depression. Am J Psychiatry
1988;145:93743.
Mayer-Kress G, Kaneko K. Spatiotemporal chaos and noise. J Stat Phys
1989;54:1489508.
Nandrino JL, Pezard L, Martinerie J, Massioui F, Renault B, Jouvent R.
Decrease of complexity in EEG as a symptom of depression.
Neuroreport 1994;5:52830.
Nikulin VV, Brismar T. Long-range temporal correlations in electroencephalographic oscillations: relation to topography, frequency band,
age and gender. Neuroscience 2005;130:54958.
Nystrom C, Matousek M, Hallstrom T. Relationships between EEG and
clinical characteristics in major depressive disorder. Acta Psychiatr
Scand 1986;73:3904.
Palus M. Nonlinearity in normal human EEG: cycles, temporal asymmetry, nonstationarity and randomness, not chaos. Biol Cybern
1996;75:38996.
Parish LM, Worrell GA, Cranstoun SD, Stead SM, Pennell P, Litt B.
Long-range temporal correlations in epileptogenic and non-epileptogenic human hippocampus. Neuroscience 2004;125:106976.
Peng CK, Buldyrev SV, Goldberger AL, Havlin S, Sciortino F, Simons M.
Long-range correlations in nucleotide sequences. Nature 1992;356:
16870.
Peng CK, Havlin S, Stanley HE, Goldberger AL. Quantication of scaling
exponents and crossover phenomena in nonstationary heartbeat time
series. Chaos 1995;5:827.
Peng CK, Hausdor JM, Goldberger AL. Fractal mechanisms in neural
control: human heart beat and gait dynamics in health and disease. In:
Walleczek J, editor. Self-organized biological dynamics and nonlinear
control. Cambridge: Cambridge University Press; 2000.
Pereda E, Gamundi A, Rial R, Gonzalez J. Non-linear behaviour of
human EEG: fractal exponent versus correlation dimension in wake
and sleep stages. Neurosci Lett 1998;250:914.
Pollock V, Schneider L. Quantitative, waking EEG research on depression. Biol Psychiatry 1990;27:75780.

Popivanov D, Mineva A. Testing procedures for non-stationarity and

non-linearity in physiological signals. Math Biosci 1999;157:30320.
Rapp PE. Chaos in the neurosciences: cautionary tales from the frontier.
Biologist 1993;40:8994.
Rapp PE, Bashore T, Martinere J, Albano A, Zimmerman I, Mess A.
Dynamics of the brain electrical activity. Brain Topogr 1989;2:99118.
Schaer CE, Davidson RJ, Saron C. Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects. Biol
Psychiatry 1983;18:75362.
Septien L, Giroud M, Didi-Roy R. Depression and partial epilepsy:
relevance of laterality of the epileptic focus. Neurol Res 1993;15:1368.
Sheline YI, Wang PW, Gado MH, Csernansky JG, Vannier MW.
Hippocampal atrophy in recurrent major depression. Proc Natl Acad
Sci USA 1996;93:390813.
Stam CJ, Montez T, Jones BF, Rombouts SARB, van der Made Y,
Pijnenburg YAL, et al. Disturbed uctuation of resting state EEG
synchronization in Alzheimers disease. Clin Neurophysiol 2005;116:
70815.
Theiler J, Rapp PE. Re-examination of the evidence for low-dimensional,
nonlinear structure in the human electroencephalogram. Electroencephalogr Clin Neurophysiol 1996;98:21322.
Thomasson N, Pezard L, Allilaire J, Renault B, Martinerie J. Nonlinear
EEG changes associated with clinical improvement in depressed
patients. Nonlinear Dynamics Psychol Life Sci 2000;4:20318.
Vakili K, Pillay SS, Lafer B, Fava M, Renshaw PF, Bonello-Cintron CM.
Hippocampal volume in primary unipolar major depression: a
magnetic resonance imaging study. Biol Psychiatry 2000;47:108790.
Varela F, Lachaux JP, Rodriguez E, Martinerie J. The brainweb: phase
synchronization and large-scale integration. Nat Rev Neurosci 2001;2:
22939.
Watters PA, Martin F. A method for establishing long-range power law
correlations from the electroencephalogram. Biol Psychiatry 2004;66:
7989.
Widlocher D. Retardation: a basic emotional response? In: Davis J, Mass
J, editors. The aective disorders. Washington, DC: American Psychiatric Press; 1983.