You are on page 1of 12

Annals of Biomedical Engineering, Vol. 36, No. 1, January 2008 ( 2007) pp.

141–152
DOI: 10.1007/s10439-007-9402-y

Regional Analysis of Spontaneous MEG Rhythms in Patients


with Alzheimer’s Disease Using Spectral Entropies
JESÚS POZA,1 ROBERTO HORNERO,1 JAVIER ESCUDERO,1 ALBERTO FERNÁNDEZ,2 and CLARA I. SÁNCHEZ1
1
Biomedical Engineering Group, Department T.S.C.I.T., E.T.S. Ingenieros de Telecomunicación, University of Valladolid,
Camino del Cementerio s/n, 47011 Valladolid, Spain; and 2Centro de Magnetoencefalografı́a Dr. Pérez-Modrego, Complutense
University of Madrid, Madrid, Spain
(Received 3 July 2007; accepted 30 October 2007; published online 10 November 2007)

Abstract—Alzheimer’s disease (AD) is the most common countries where an increase in life expectancy is
form of dementia. Ageing is the greatest known risk factor expected.7 Early symptoms of AD include memory loss
for this disorder. Therefore, the prevalence of AD is expected and concentration problems, while as the dementia
to increase in western countries due to the rise in life
expectancy. Nowadays, a low diagnosis accuracy is reached, progresses AD patients may suffer aphasia, apraxia,
but an early and accurate identification of AD should be and agnosia, accompanied with general cognitive
attempted. In this sense, only a few studies have focused on symptoms such as confusion, personality and behavior
the magnetoencephalographic (MEG) AD patterns. This changes, impaired judgment, and disorientation.6
work represents a new effort to explore the ability of three Brain changes in AD are related to the appearance
entropies from information theory to discriminate between
spontaneous MEG rhythms from 20 AD patients and 21 of two characteristic lesions: senile or neuritic plaques
controls. The Shannon (SSE), Tsallis (TSE), and Rényi and neurofibrillary tangles. They both are formed by
(RSE) spectral entropies were calculated from the time- clusters of proteins accumulated in two specific brain
frequency distribution of the power spectral density (PSD). regions: the hippocampus and the cerebral cortex.
The entropies provided statistically significant lower values Whereas the small clumps of beta-amyloid may induce
for AD patients than for controls in all brain regions
(p < 0.0005). This fact suggests a significant loss of irregu- a degeneration of synapses, the neurofibrillary tangles
larity in AD patients’ MEG activity. Maximal accuracy of destroy a vital cell transport system made of proteins.11
87.8% was achieved by both the TSE and RSE (90.0%, Several potential biomarkers for AD have been
sensitivity; 85.7%, specificity). The statistically significant proposed. Among them, the levels Ab1-42 and isoforms
results obtained by both the extensive (SSE and RSE) and of tau protein in cerebrospinal fluid (CSF) have been
non-extensive (TSE) spectral entropies suggest that AD
could disturb long and short-range interactions causing an extensively reported in the literature.7 Many other fluid
abnormal brain function. biomarkers have been analyzed in both serum and
CSF, including isoprostanes and glycation, inflamma-
Keywords—Magnetoencephalogram, Time-frequency distri- tion and oxidative stress markers.44 Various plasma
bution, Shannon spectral entropy, Tsallis spectral entropy, markers are also being investigated, including amyloid
Rényi spectral entropy. peptide proteins and similar molecules to those ana-
lyzed in CSF.44 Structural neuroimaging studies offer
complementary advantages to the biological markers.
INTRODUCTION Thus, positron emission tomography (PET) imaging of
glucose metabolism and magnetic resonance imaging
Alzheimer’s disease (AD) is a progressive neurode- (MRI) volumetry of medial temporal lobe have shown
generative disorder of unknown etiology. It is the promising results as diagnostic screening in clinical
leading form of dementia representing about 50–60% trials.26 Nevertheless, their routinely use is not rec-
of all cases.7 Several risk factors have been identified.7 ommended in any consensus guidelines for diagnosis of
However, the greatest known risk factor for AD is the disease.7
ageing, which is especially important in western Clinical diagnosis of AD is based on a complete
medical history together with physical, psychiatric, and
neurological examinations.25,34 Additionally, labora-
Address correspondence to Jesús Poza, Biomedical Engineering
tory studies, such as thyroid-function tests and
Group, Department T.S.C.I.T., E.T.S. Ingenieros de Telecom-
unicación, University of Valladolid, Camino del Cementerio s/n, assessment of vitamin B12 deficiency, are recom-
47011 Valladolid, Spain. Electronic mail: jespoz@tel.uva.es mended to identify other forms of dementia and
141
0090-6964/08/0100-0141/0  2007 Biomedical Engineering Society
142 POZA et al.

coexisting disorders associated with ageing.25 Struc- comparison with controls suggesting a loss of func-
tural neuroimaging techniques such as computer tional connectivity.38 In a recent study analyzing
tomography (CT) or MRI are also appropriate to MEG recordings at rest, it has been shown that
exclude other causes of dementia.25 Nevertheless, the functional connectivity in AD is characterized by
diagnosis accuracy is low, with sensitivity of around specific changes of long and short distance interac-
81% and specificity of 70%.25 Hence, a definite con- tions in several frequency bands.39 Decreased levels of
firmation of AD can be only made by post-mortem functional connectivity indicate that AD is related to
examination of brain tissue.7,34 an abnormal function of the large scale brain net-
In spite of the relatively low diagnosis accuracy, an works. Nevertheless, further studies should confirm
early and accurate identification of AD should be this hypothesis.
attempted. Current therapies are more effective if the We conclude that MEG signals stem from a com-
medication is administered in the initial stages of the plex system where both short and long-range (for in-
disease.11 Additionally, an early diagnosis enables to terneural distances) phenomena are simultaneously
develop strategies for coping with the disease. Elec- present. Information measures have recently demon-
troencephalographic (EEG) brain activity has been strated to be appropriate to characterize complex
extensively analyzed to obtain early markers of AD, recordings. Previous studies have successfully applied
while only a few studies have focused on the magne- several quantifiers based on extensive (e.g., Shannon
toencephalographic (MEG) disease patterns.23,37 Al- and Rényi entropies) and non-extensive (e.g., Tsallis
though EEG and MEG recordings are generated by entropy) information measures to analyze EEG sig-
synchronous oscillations of pyramidal neurons, they nals.2,5,9,24,33,40,41 While extensive quantifiers, such as
reflect slightly different characteristics.20 The EEG Shannon and Rényi entropies, have proved successful
measures the electric fields induced by all primary in describing systems with short-range interactions,
currents, whereas the MEG is sensitive only to current non-extensive measures, such as Tsallis entropy, are
flows oriented parallel to the scalp.20,21 Furthermore, able to describe a system where the effective interac-
the EEG is strongly influenced by a wide variety of tions are of long-range.5,40 In addition, MEG record-
factors such as head structures (e.g., skull and extra- ings, like EEG signals, are non-stationary and their
cerebral tissues) or several technical and methodolog- characteristics may change over time. Non-stationary
ical issues (e.g., electrode placement, reference point or analysis techniques, such as time-frequency distribu-
even adhesive compound of the electrode to the skin). tions, are appropriate to accurately describe their
MEG recordings are less distorted than EEG signals properties. Previous efforts have exploited the analogy
on the scalp due to the insensitivity of magnetic fields between power spectral densities and probability den-
to inhomogeneities in the head.21 In addition, MEG sities to apply several entropic forms to time-frequency
signals are reference-free recordings and offer higher distributions.2,3
spatial resolution than conventional EEG.20 The present study is a new approach to explore the
A few studies have found several abnormalities in ability of several measures derived from information
spontaneous MEG recordings of AD patients when theory to characterize MEG rhythms in AD. This
they were compared to those from healthy controls. work has been developed on the basis of the results
MEG studies using relative power have shown in- obtained in a pilot study where we tested several
creased slow rhythms and reduced fast activity in entropies using the averaged relative power over all
AD.4,14,28 This issue has been also observed by means sensors as a probability distribution.30 Our findings
of spectral quantifiers such as mean frequency,13,31 suggested both a slowing of MEG rhythms and a loss
individual alpha peak28,29,31 and transition fre- of irregularity in AD. Further work should be devel-
quency.31 Some authors have found a global decrease oped to obtain a temporal evolution of the quantifiers,
in irregularity31 and complexity19 in AD patients’ to assess the influence of the entropic index and to
MEG recordings at rest. However, the loss of com- perform a spatial analysis. The parameters used in the
plexity has been only described in the high frequencies present study are extensive and non-extensive entropies
when a detailed spectral analysis in several frequency based on the time-frequency distribution, which mea-
bands was performed.8 From another point of view, sure the irregularity or disorder of MEG recordings.
other approaches have focused on analyzing functional Initially, the Fourier transform (FT) was used to
connections rather than local abnormalities in AD. compute the power spectral density (PSD) into non-
Their findings showed both a decrease of coherence overlapping temporal windows for each MEG
values in the alpha band18 and a general decrease of recording from AD patients and controls. Shannon
coherence in all frequency bands.4 A reduced level of (SSE), Tsallis (TSE), and Rényi (RSE) spectral
synchronization has also been reported in the upper entropies were then calculated to explore the irregu-
alpha, beta, and gamma bands of AD patients in larity of the MEG signals. Additionally, both the RSE
Entropy Analysis of MEG Rhythms in Alzheimer’s Disease 143

extensive and the TSE non-extensive entropies were We did not obtain significant differences (p > 0.05,
parameterized by the entropic index q, which was Student’s t-test) when the mean age of AD patients
modified to emphasize particular characteristics of the and controls was compared. In addition, all healthy
associated dynamics. volunteers and patients’ caregivers accepted to partic-
In ‘‘Experimental setup’’ section, the demographic ipate in the study and gave written informed consent.
and neuropsychological data of the study population The trial was approved by the study center’s Research
are presented, together with the characteristics of the Ethics Committee.
MEG recordings. Section ‘‘Methods’’ briefly summa-
rizes the definitions of Shannon, Tsallis, and Rényi
MEG Recording
entropies and their application on time-frequency
plane. Results are presented in ‘‘Results’’ section, MEG signals were obtained using a 148-channel
where the role of the parameter q is analyzed and whole-head magnetometer (MAGNES 2500 WH, 4D
regional entropy analyses are performed. In ‘‘Discus- Neuroimaging), which was confined in a magnetically
sion’’ section the relevant results are discussed, while shielded room. During the data acquisition, subjects
‘‘Conclusions’’ section summarizes the major contri- were asked to remain relaxed, awake and with eyes
butions of the paper. closed in order to minimize the presence of artifacts in
the recordings. Additionally, MEG signals were con-
tinuously monitored to prevent drowsiness. Five min-
utes of spontaneous MEG activity were acquired with
EXPERIMENTAL SETUP a sample frequency of 678.17 Hz. Initially, both a 0.1–
200 Hz hardware band-pass filter and a 50 Hz notch
Selection of Subjects
filter were applied. Then, each MEG recording was
We analyzed MEG signals from 41 subjects re- downsampled by a factor of four to reduce the
corded in the ‘‘Centro de Magnetoencefalografı́a data length. Artifact-free epochs of length 10 s
Dr. Pérez-Modrego’’ of the Complutense University (26.43 ± 5.49 artifact-free epochs per channel and
of Madrid (Spain). Twenty subjects (7 men and 13 subject, mean ± SD) were selected for further analy-
women, age = 73.05 ± 8.65 years, mean ± standard sis. Prior to calculate spectral entropies, MEG signals
deviation SD) were AD patients derived from the were processed using a 560th order finite impulse re-
‘‘Asociación de Familiares de Enfermos de Alzheimer’’ sponse (FIR) filter designed with a Hamming window
(AFAL) and the Geriatric Unit of the ‘‘Hospital Clı́- and cut-off frequencies at 0.5 and 70 Hz. The selected
nico Universitario San Carlos’’ from Madrid. They frequency range enables to keep the relevant spectral
were diagnosed on the basis of exhaustive medical, content and to minimize the presence of oculographic
physical, neurological, psychiatric, and neuropsycho- and myographic artifacts.
logical examinations, which were complemented with
brain scans to exclude other causes of dementia. All of
them fulfilled the criteria for probable AD according to METHODS
the clinical guidelines of the National Institute of
Neurological and Communicative Disorders and Definition of Spectral Entropies
Stroke and the AD and Related Disorders Association The FT is a well-known technique used to analyze
(NINCDS-ADRDA).27 Both the Mini-Mental State the spectral content of a signal. A common method to
Examination (MMSE)17 and the Functional Assess- assess the distribution of the power as a function of
ment Staging (FAST)32 were used to evaluate the frequency is the PSD. In the present work, the PSD
cognitive function. AD patients obtained mean scores was calculated from the FT of the autocorrelation
of 17.85 ± 3.91 and 4.00 ± 0.32 on the MMSE and function.13,14,31 Then, the spectral content between 0.5
FAST, respectively. None of the AD patients suffered and 70 Hz was selected and the PSD was normalized
from any other significant medical, neurological, and to a scale from 0 to 1 leading to the normalized PSD
psychiatric disorder. They were not taking any drug (PSDn)
that could affect the MEG recordings.
Twenty-one healthy subjects (9 men and 12 women, PSDðfÞ
PSDn ðfÞ ¼ : ð1Þ
age = 70.29 ± 7.07 years, mean ± SD) were en- P
70Hz
rolled in the study as control group. They were cog- PSDðfÞ
f¼0:5Hz
nitively normal elderly controls with no history of
neurological or psychiatric disorders. The mean After the normalization, it follows that
P70 Hz
MMSE and FAST scores were 29.10 ± 1.00 and f¼0:5 Hz PSD n ðfÞ ¼ 1: The PSDn can be considered
1.71 ± 0.46, respectively. as a probability distribution with N points, whose
144 POZA et al.

definition will be further extended to the time-fre- it is an additive entropy, which is based on a loga-
quency plane. This representation provides a suitable rithmic transformation of a given probability distri-
tool for analyzing and characterizing the power spec- bution. The PSDn represents again the probability
trum, since it is possible to apply several entropies density function, analogously to previous EEG stud-
based on a probability density function. Thus, in the ies.2,24 Hence, the Rényi entropy can be considered as
present work three entropies were calculated from the an alternative way to estimate the irregularity of time-
PSDn: the SSE, TSE, and RSE. frequency distributions.3 This entropic form is
Shannon’s entropy is a disorder quantifier whose parameterized by an entropic index q 2 <, in order that
original meaning implies uncertainty of information in the Rényi entropy can be reduced to the Boltzmann–
terms of disorder, discrepancy and diversity.5 The Gibbs entropy in the limit q fi 1.35 The definition of
definition of Shannon’s entropy is based on a proba- the RSE reads
bility distribution. Previous EEG and MEG studies ( )
7X
0Hz
have used the PSDn to compute the SSE as a statistical 1 q
RSEðqÞ ¼ ln ½PSDn ðfÞ ; q>0: ð4Þ
descriptor of the signal’s irregularity in AD.1,31 The 1q f¼0:5Hz
irregularity is estimated in terms of the flatness of the
PSDn.36 A uniform PSDn with a broad spectral con- The previous entropies (SSE, TSE, and RSE) were
tent (e.g., a highly irregular signal like white noise) normalized in order to take values in the 0–1 interval.
gives a high SSE value. On the contrary, a narrow Hence, the aforementioned entropic forms were divided
PSDn with only a few spectral components (e.g., a by their corresponding maximum possible values. They
highly predictable signal like a sum of sinusoids) yields were ln(N) for both the SSE and the RSE and [1 - N (1-
q)
a low SSE value. In the present work, the definition of ] for the TSE, where N was the number of frequency
the SSE is based on Shannon’s entropy computed over bins and q represented the entropic index.5,33
the PSDn from 0.5 to 70 Hz22 and it is written by
X
70Hz
SSE ¼  PSDn ðfÞ  ln ½PSDn ðfÞ: ð2Þ Time Evolution of Spectral Entropies
f¼0:5Hz The MEG recordings irregularity can be analyzed
The second entropic form is the Tsallis entropy. It is indirectly applying several entropy definitions to their
a generalized information measure, which extends the power frequency distributions. In this way, entropy
notion of Shannon’s entropy.42 The Tsallis entropy is a provides a description of average uncertainty for a
non-logarithmic entropy, which can be useful to ex- given signal.41 However, MEGs are non-stationary
plore the properties of a probability distribution from recordings and their properties can be accurately de-
a new mathematical framework in relation to Shan- scribed by using non-stationary signal analysis tech-
non’s entropy.42,43 Previous EEG studies have suc- niques, such as time-frequency distributions.2,3 In the
cessfully used the relative power distribution of wavelet present work, an alternative entropic form based on
coefficients to analyze different brain states in epi- the sliding temporal window technique was applied to
lepsy.9,33 Given the fact that the Tsallis entropy is a obtain the entropy evolution.2,3 Each MEG epoch of
non-extensive measure, it is able to characterize a 10 s (M = 1696 samples) was divided into non-over-
system where long-range interactions appear.5 The lapping temporal fragments of 0.5 s, whose length is
degree of non-extensivity is stated by the entropic in- denoted as L (L = 84 samples) and each time interval
dex q 2 <.12 The parameter q might be considered as a is identified by i (i = 1,..., NT, with NT = M/L). The
zoom lens, which can be focused on long-range inter- PSD is calculated for each temporal window and the
actions (low q values) or on short abrupt changes (high PSDn is computed from the PSD
q values).43 The Tsallis entropy recovers the definition PSDðiÞ ðfÞ
of the Boltzmann–Gibbs entropy in the limit q fi 1.42 PSDðniÞ ðfÞ ¼ ; i ¼ 1; . . . ; NT : ð5Þ
P
70Hz
The definition of the discrete version of the TSE is PSDðiÞ ðfÞ
given by f¼0:5Hz

1 7X
0Hz The evolution of the SSE, TSE, and RSE is com-
ðiÞ
TSEðqÞ ¼ fPSDn ðfÞ  ½PSDn ðfÞq g; q>0: puted from the PSDn ðfÞ and each value is assigned to
q  1 f¼0:5Hz
the central point of the corresponding interval
ð3Þ 7X
0Hz h i
SSEðiÞ ¼  PSDðniÞ ðfÞ  ln PSDðniÞ ðfÞ ; i ¼ 1;. ..;NT
Finally, the Rényi entropy is a generalized infor-
f¼0:5Hz
mation measure, which also extends the definition of
Shannon’s entropy.35 Similarly to Shannon’s entropy, ð6Þ
Entropy Analysis of MEG Rhythms in Alzheimer’s Disease 145

X
70 Hz n h iq o measures analyses. Differences were considered statis-
ðiÞ 1
TSE ðqÞ ¼ PSDðniÞ ðfÞ  PSDðniÞ ðfÞ ; tically significant for p < 0.05.
q  1 f¼0:5 Hz The classification performance of each parameter
i ¼ 1; . . . ; NT was evaluated using a linear discriminant analysis
(LDA) with a leave-one-out cross-validation proce-
ð7Þ dure. Classification statistics were shown in terms of
( ) sensitivity (percentage of AD patients with a correct
1 X
70 Hz h iq diagnosis), specificity (proportion of controls properly
ðiÞ ðiÞ
RSE ðqÞ ¼ ln PSDn ðfÞ ;
1q ð8Þ recognized) and accuracy (total fraction of AD pa-
f¼0:5 Hz
tients and healthy subjects well classified).
i ¼ 1; . . . ; NT : All statistical analyses were performed using SPSS
software (version 14.0; SPSS Inc, Chicago, Ill).
Finally, the previous quantifiers are normalized by
their maximum possible values, which were specified in
the previous section.
RESULTS
To characterize the entropy evolution by a single
value, the temporal average per parameter and subject Optimization of the Entropic Index q
was calculated for each quantifier
In a first stage, we analyzed the role of the entropic
1 X NT
index q in the parameterized entropies: TSE and RSE.
hSSEi ¼  SSEðiÞ ð9Þ
NT i¼1 The evolution of the TSE and RSE over each of the
10 s MEG segments was computed taking non-over-
lapping temporal windows of 0.5 s. Results were
1 X NT
averaged for each channel and each subject to obtain
hTSEðqÞi ¼  TSEðiÞ ðqÞ ð10Þ
NT i¼1 an entropy evolution per channel and subject. Finally,
temporal averages for the TSE and the RSE were
1 X NT computed to obtain a quantitative measure per subject
hRSEðqÞi ¼  RSEðiÞ ðqÞ: ð11Þ and q value. The dependence of the TSE and RSE on
NT i¼1
the entropic index q was explored modifying its values
All calculations were carried out with the software (q = {0.25, 0.5,...,5}) according to previous EEG
package Matlab (version 7.0; Mathworks, Natick, MA). studies.2,3,9,24,33
The TSE showed a significant main effect of group
(F(1,38) = 27.909; p = 0.000005) and a significant
Statistical Analysis group by q interaction (F(19,722) = 13.792;
p = 0.000514). Figure 1 depicts the evolution of the
Both the Kolmogorov–Smirnov and the Shapiro– p-values from the univariate ANOVAs with contrasts
Wilk tests were used to evaluate the normal distribu- and age as a covariate in function of the entropic index
tion of each variable, while homoscedasticity was q, when the ÆTSEæ averaged over all channels is com-
assessed with Levene’s test. After the descriptive pared between AD patients and controls. As it can be
analysis, log-transformed variables met parametric test seen, the TSE with both low and high q values
assumptions. Firstly, the mean entropy values aver- achieved the greatest p-values. The most significant
aged over all channels were analyzed using two-way
repeated measures ANOVAs (with group as between-
subject factor and parameter q as within-subject fac-
tor) with age as a covariate. Secondly, optimal mean
entropy values at five brain regions (anterior, central,
left lateral, posterior, and right lateral) were compared
between AD patients and controls by means of two-
way repeated measures ANOVAs (with group as be-
tween-subject factor and brain region as within-subject
factor) with age as a covariate. Finally, univariate
ANOVAs with contrasts and age as a covariate were
performed when previous analyses showed significant
interactions. In order to correct possible violations of
FIGURE 1. Evolution of the statistical significance in func-
the sphericity assumption and to reduce type I errors, tion of the entropic index q when the mean TSE over all
Greenhouse-Geisser epsilon was used in all repeated channels is compared between AD patients and controls.
146 POZA et al.

FIGURE 2. Evolution of the statistical significance in func-


tion of the entropic index q when the mean RSE over all
channels is compared between AD patients and controls.

differences (p = 0.000001) were obtained by the


TSE(q = 2).
The RSE exhibited a significant main effect of group
(F(1,38) = 34.189; p = 0.000001) and a significant FIGURE 3. Illustration of sensor grouping into the five brain
group by q interaction (F(19,722) = 7.794; regions considered for topographic analyses: anterior, cen-
p = 0.006891). Figure 2 illustrates the evolution of the tral, left lateral, posterior and right lateral.
p-values from the univariate ANOVAs with contrasts
and age as a covariate in function of the entropic index of the spectrum flatness. It is illustrated in Fig. 4,
q, when the ÆRSEæ averaged over all channels is com- where the mean SSE evolution over a 10 s period for
pared between AD patients and controls. The statisti- each brain region in both groups is shown. Although
cal significance achieved with the RSE augmented as the most significant results were obtained in the right
q increased. It can be observed that this trend is lateral region (p-value = 0.000004; 80.5%, accuracy;
stable around q = 3. The most significant result 80.0%, sensitivity; 81.0%, specificity), it is noteworthy
(p < 0.000001) was obtained with RSE(q = 3.5). that the statistical analyses showed similar statistical
differences and classification parameters in the left
lateral (p-value = 0.000019; 80.5%, accuracy; 75.0%,
Regional Entropy Analysis
sensitivity; 85.7%, specificity) and anterior areas (p-
In a second step, we explored the regional patterns value = 0.000053; 82.9%, accuracy; 85.0%, sensitiv-
of the SSE, as well as the previously optimized ity; 81.0%, specificity). In addition, the SSE displayed
TSE(q = 2) and RSE(q = 3.5). The evolution of the a higher variability over the 10 s interval in AD
previous quantifiers over each of the 10 s MEG seg- patients than in healthy subjects.
ments was computed taking non-overlapping temporal The TSE(q = 2) exhibited a significant main effect
windows of 0.5 s. According to Fig. 3, five brain areas of group (F(1,38) = 35.269; p = 0.000001) and a
(anterior, central, left lateral, posterior, and right lat- significant group by region interaction (F(4,152) =
eral) were defined to obtain a regional entropy analy- 4.372; p = 0.013079). Table 1 shows that AD patients
sis. Results were averaged in each of the five brain obtained a significantly lower TSE(q = 2) than con-
regions, obtaining an entropy evolution per region trols in all brain regions. This result can be seen in
and subject. Additionally, temporal averages for the Fig. 5, which depicts the mean TSE(q = 2) evolution
entropies were computed to obtain a quantitative over a 10 s period for each brain region in both groups,
measure per region and subject. suggesting again a decrease in the irregularity of AD
The SSE showed a significant main effect of group patients’ MEGs. The inspection of Tables 1 and 2 re-
(F(1,38) = 25.986; p = 0.000010), whereas no signifi- veals that the most significant differences (p-value =
cant group by region interaction was observed 0.000001) were achieved in the right lateral area, with
(p = 0.072579). Results of univariate ANOVAs with an accuracy of 78.0% (75.0%, sensitivity; 81.0%,
contrasts and age as a covariate are shown in Table 1. specificity). Similar statistical results were obtained in
Classification analyses using LDA with a leave-one-out the left lateral (p-value = 0.000003) and anterior re-
cross-validation procedure are shown in Table 2. AD gions (p-value = 0.000003). Nevertheless, in the left
patients obtained significantly lower SSE values than lateral and anterior areas the accuracy reached 80.5%
controls in all cerebral regions, which suggests an (75.0%, sensitivity; 85.7%, specificity) and 87.8%
irregularity decrease in AD patients’ MEGs, in terms (90.0%, sensitivity; 85.7%, specificity), respectively.
Entropy Analysis of MEG Rhythms in Alzheimer’s Disease 147

TABLE 1. Mean SSE, TSE(q = 2) and RSE(q = 3.5) values averaged in each brain region for each group and the corresponding
p-values from ANOVA with contrasts and age as a covariate.

Controls AD patients
Region Parameter Mean ± SD Mean ± SD p-value

Anterior hSSE i 0.833 ± 0.003 0.720 ± 0.005 0.000053


hTSEðq ¼ 2Þi 0.969 ± 0.001 0.916 ± 0.003 0.000003
hRSEðq ¼ 3:5Þi 0.714 ± 0.006 0.524 ± 0.007 0.000003
Left lateral hSSE i 0.818 ± 0.003 0.723 ± 0.004 0.000019
hTSEðq ¼ 2Þi 0.969 ± 0.001 0.928 ± 0.002 0.000003
hRSEðq ¼ 3:5Þi 0.698 ± 0.006 0.556 ± 0.005 0.000003
Central hSSE i 0.859 ± 0.001 0.786 ± 0.004 0.000172
hTSEðq ¼ 2Þi 0.979 ± 0.000 0.951 ± 0.002 0.000022
hRSEðq ¼ 3:5Þi 0.760 ± 0.003 0.629 ± 0.006 0.000008
Right lateral hSSE i 0.823 ± 0.003 0.722 ± 0.005 0.000004
hTSEðq ¼ 2Þi 0.969 ± 0.001 0.928 ± 0.003 0.000001
hRSEðq ¼ 3:5Þi 0.702 ± 0.005 0.556 ± 0.007 0.000000
Posterior hSSE i 0.815 ± 0.002 0.735 ± 0.006 0.000137
hTSEðq ¼ 2Þi 0.968 ± 0.001 0.937 ± 0.004 0.000102
hRSEðq ¼ 3:5Þi 0.692 ± 0.005 0.577 ± 0.007 0.000023

SD: Standard deviation; p-value: Statistical significance.

TABLE 2. Diagnostic test results derived from linear were obtained in the right lateral region with an
discriminant analysis with a leave-one-out cross-validation accuracy of 78.0% (75.0%, sensitivity; 81.0%, speci-
procedure for the mean SSE, TSE(q = 2) and RSE(q = 3.5)
averaged in each brain region. ficity). Similar statistical results and higher classifica-
tion parameters were found in the anterior (p-
Sensitivity Specificity Accuracy value = 0.000003; 87.8%, accuracy; 90.0%, sensitiv-
Region Parameter (%) (%) (%)
ity; 85.7%, specificity), left lateral (p-value =
Anterior hSSE i 85.0 81.0 82.9 0.000003; 82.9% accuracy; 80.0%, sensitivity; 85.7%,
hTSEðq ¼ 2Þi 90.0 85.7 87.8 specificity) and central areas (p-value = 0.000008;
hRSEðq ¼ 3:5Þi 90.0 85.7 87.8 82.9%, accuracy; 75.0%, sensitivity; 90.5%, specificity)
Left lateral hSSE i 75.0 85.7 80.5 than in the right lateral region.
hTSEðq ¼ 2Þi 75.0 85.7 80.5
hRSEðq ¼ 3:5Þi 80.0 85.7 82.9
Central hSSE i 70.0 76.2 73.2 DISCUSSION
hTSEðq ¼ 2Þi 65.0 85.7 75.6
hRSEðq ¼ 3:5Þi 75.0 90.5 82.9 We explored the ability of several measures derived
Right lateral hSSE i 80.0 81.0 80.5 from information theory to discriminate between
hTSEðq ¼ 2Þi 75.0 81.0 78.0 spontaneous MEG rhythms of 20 AD patients and 21
hRSEðq ¼ 3:5Þi 75.0 81.0 78.0 control subjects. We calculated extensive (Shannon
Posterior hSSE i 65.0 81.0 73.2 and Rényi) and non-extensive (Tsallis) entropies based
hTSEðq ¼ 2Þi 70.0 76.2 73.2 on the time-frequency plane2,3 to measure the irregu-
hRSEðq ¼ 3:5Þi 75.0 85.7 80.5 larity of MEGs in AD. Initially, we analyzed the
dependence of both the TSE and RSE on the entropic
Analogously to the SSE, the TSE showed a higher index q for discriminating between AD patients and
variability over the 10 s interval in AD patients than in controls. We found q = 2 and q = 3.5 as optimal
healthy subjects. entropic indexes for the TSE and RSE, respectively.
Finally, the RSE(q = 3.5) exhibited a significant Detailed regional analyses were then performed using
main effect of group (F(1,38) = 35.440; p = 0.000001) the SSE, TSE(q = 2) and RSE(q = 3.5) in order to
and a significant group by region interaction explore their spatial and temporal patterns in AD.
(F(4,152) = 4.771; p = 0.006325). Table 1 indicates MEGs from AD patients had significantly lower en-
that AD patients obtained significantly lower tropy values than those from controls in all brain
RSE(q = 3.5) values than controls in all brain regions. regions, which suggests that AD is accompanied by an
The statistically significant decrease in the irregularity overall irregularity decrease.
of AD patients’ MEG recordings is shown in Fig. 6. The dependence analysis of the entropies on the
The most significant differences (p-value < 0.000001) parameter q revealed that the TSE and RSE exhibited
148 POZA et al.

FIGURE 4. Grand-average of the SSE evolution over a 10 s period for AD and control groups in each brain region: anterior (A),
central (C), left lateral (L), posterior (P), and right lateral (R).

different behaviors. Similarly, previous EEG studies gradually increased. However, our results indicate that
have observed that quantifiers based on Rényi this trend appears with q values greater than 2 or 3.5
entropy are less sensitive to variations of the parameter by the TSE and RSE, respectively. Differences may be
q (q > 1) due to its logarithmic form,33,41 whereas due to the fact that we computed both the Tsallis and
measures based on the Tsallis statistic showed a strong Rényi entropies from a time-frequency distribution,
dependence on the entropic index.41 This result can whereas the aforementioned works used an alternative
also be observed in the present work. Entropic indexes time-dependent entropy definition based on the sliding
greater than 1 produced changes in the significance temporal window technique.10
level more important with the TSE than with the RSE. Regional analyses of the entropy patterns showed
Figures 1 and 2 show that the ability of the TSE to significant (p < 0.0005) lower SSE, TSE(q = 2) and
accurately detect AD is maximal using values of q close RSE(q = 3.5) values in AD patients’ MEGs than in
to 2, while the statistical differences between groups controls for all cerebral regions. The SSE decrease in
are gradually reduced as q either decreases or increases AD suggests that the disease is accompanied by a loss
with respect to the previous q value. Regarding to the of frequency components, which implies a decrease of
RSE, the statistical differences between AD patients irregularity with respect to the flatness of the power
and controls gradually increased as q augmented. spectrum. The TSE(q = 2) and RSE(q = 3.5) evolu-
Nevertheless, entropic indexes higher than 3.5 did not tions reveal again that the disorder was reduced in AD
offer significant improvement to distinguish between patients compared to controls, supporting the loss of
groups. The ability of q to enhance the accuracy to irregularity in AD. Previous MEG studies have also
identify particular brain states has been previously reported an overall decrease in both irregularity31 and
discussed. EEG studies detecting epileptic seizures and complexity19 in AD using the SSE and the Lempel–Ziv
characterizing brain function by means of Tsallis and complexity, respectively. On the contrary, van Cap-
Rényi entropies have reported a decrease in the pellen van Walsum et al.8 only described a complexity
‘‘background EEG-detection-power’’ as the parameter loss in the high frequencies when a detailed spectral
q (q > 1) increased.5,9,41 We have observed that the analysis in several frequency bands was performed
ability to discriminate between AD patients and con- using the neural complexity. An advantage of the
trols did not improve when the entropic index was methods used in this work, compared to the previous
Entropy Analysis of MEG Rhythms in Alzheimer’s Disease 149

FIGURE 5. Grand-average of the TSE(q = 2) evolution over a 10 s period for AD and control groups in each brain region: anterior
(A), central (C), left lateral (L), posterior (P), and right lateral (R).

efforts, is to take into account both temporal and measures) was used to characterize the PSD of MEGs
spatial characteristics of the MEG signals. This is due in AD.31 However, it is noteworthy that the classifi-
to the fact that the quantifiers were applied to the time- cation analyses were performed using a LDA without a
frequency plane. We have detected not only lower leave-one-out cross-validation procedure. As it was
entropy values, but also a more irregular entropy previously mentioned, the TSE and RSE have been
evolution in AD patients than in controls. Addition- used in the past to characterize long and short-range
ally, the decreasing entropy values should be inter- interactions in complex systems, respectively.5,40 Our
preted from a spectral point of view. We have observed findings indicate that both entropic formalisms showed
that AD implies a loss of spectral components, which statistically significant results. Therefore, we can
can be associated with a reduction of information hypothesize that the brain magnetic activity in AD
content.3 Some authors claimed that the change in may be generated by a system where abnormal long
information entropy within the EEG may reflect a real and short-range interactions appear simultaneously.
variation in cortical functional organization.24 Given This issue has been previously suggested by Stam
the fact that entropy is usually identified with an et al.,39 who found that the functional connectivity of
indicator of the system’s disorder, it has been spontaneous MEG activity in AD was characterized by
hypothesized that a reduction in entropy may imply specific changes of both long and short distance
an information processing decrease at the cerebral interactions in several frequency bands.
cortex.24 The most significant differences in the entropy
Further inspection of the regional entropy analyses measures were found in the lateral and anterior regions.
shows that both the TSE(q = 2) and RSE(q = 3.5) However, similar p-values were also obtained in both
achieved higher significant levels (p < 0.0001) than the central and posterior regions. Berendse et al.4 analyzed
SSE (p < 0.0005) in all brain regions. Moreover, the the absolute power of spontaneous MEG recordings of
generalized entropies reached greater classification AD patients and controls. They reported significant
accuracies (between 73% and 87.8%) than the SSE changes in the power patterns over the whole head,
(between 73.2% and 82.9%). A similar result was which can be summarized as a significant increase of the
obtained in a previous study (p = 0.00023; 82.93% low frequency power over the frontal and central areas
accuracy), where the SSE (among other spectral and a significant decrease of the high frequency power
150 POZA et al.

FIGURE 6. Grand-average of the RSE(q = 3.5) evolution over a 10 s period for AD and control groups in each brain region:
anterior (A), central (C), left lateral (L), posterior (P), and right lateral (R).

over the occipital and temporal regions.4 MEG source negative) errors should be made. The reduced number
analyses have also found differences over the different of subjects enrolled in the study implies an increase of
brain regions in AD. Fernández et al.15,16 described an beta (probability of making a type II error) and a
increased dipole density at the delta and theta bands decrease of the power of the test (i.e., 1 - beta).
over the parietal and temporal cortices in AD. On the Therefore, it should be appropriate to increase the
other hand, Osipova et al. observed that AD patients number of subjects to both minimize type II errors and
showed an increased activation of higher theta band augment the statistical power.
sources in the right temporal region,29 together with a
parieto-occipital deficit.28
Finally, some limitations of the study should be CONCLUSIONS
mentioned. It might be possible that the irregularity
decrease in AD could be drug-related. However, none This study was performed as a new approach to
of the patients were receiving any medication which analyze the ability of information theory methods to
could affect spontaneous MEG activity. On the other characterize spontaneous MEG activity from AD pa-
hand, the irregularity patterns in other neurodegener- tients and controls. Only a few MEG studies have
ative diseases, which have shown alterations in the explored the irregularity and complexity patterns in
spectrum similar to those observed in AD, should be AD.8,19,31 The parameters presented in this paper
explored. Further investigation should be attempted to combine three entropies from information theory with
extend the analyses in mild cognitive impairment, time-frequency signal processing tools. Our findings
vascular dementia, Lewy body dementia, major support the notion that AD involves an overall loss of
depression, dementia associated with Parkinson’s dis- irregularity in the electromagnetic brain activity. In
ease, Pick’s disease, Huntington’s chorea and pro- addition, the statistically significant results obtained by
gressive supranuclear palsy. both the extensive (SSE and RSE) and non-extensive
Other limitation of the work is related to the small (TSE) spectral entropies suggest that AD could disturb
sample size. Considerations about type I (probability long and short-range interactions leading to an
of a false positive) and type II (probability of a false abnormal brain function.
Entropy Analysis of MEG Rhythms in Alzheimer’s Disease 151
9
Future efforts will be addressed to explore other Capurro, A., L. Diambra, D. Lorenzo, O. Macadar, M. T.
measures from information theory to characterize Martin, C. Mostaccio, A. Plastino, J. Pérez, E. Rofman,
M. E. Torres, and J. Velluti. Human brain dynamics: the
MEG rhythms in AD. Additionally, further work
analysis of EEG signals with Tsallis information measure.
should be attempted to increase the sample size and to Physica A 265:235–254, 1999.
extend the results to other dementias. 10
Capurro, A., L. Diambra, D. Lorenzo, O. Macadar, M. T.
In summary, our findings suggest that the extensive Martin, C. Mostaccio, A. Plastino, E. Rofman, M. E.
(SSE and RSE) and non-extensive (TSE) spectral Torres, and J. Velluti. Tsallis entropy and cortical
dynamics: the analysis of EEG signals. Physica A 257:149–
entropies may lead to a better understanding of the
155, 1998.
underlying brain dynamics in AD. Furthermore, these 11
Cummings, J. L. Alzheimer’s disease. N. Engl. J. Med.
new approaches extend the concept of irregularity and 351:56–67, 2004.
12
can provide useful descriptors of the spontaneous Di Sisto, R. P., S. Martinez, R. B. Orellana, A. R. Plastino,
MEG rhythms in AD. and A. Plastino. General thermostatistical formalisms,
invariance under uniform spectrum translations, and
Tsallis q-additivity. Physica A 265:590–613, 1999.
13
Fernández, A., R. Hornero, A. Mayo, J. Poza, P. Gil-
Gregorio, and T. Ortiz. MEG spectral profile in Alzhei-
ACKNOWLEDGMENTS mer’s disease and mild cognitive impairment. Clin. Neuro-
physiol. 117:306–314, 2006.
This work has been partially supported by the grant 14
Fernández, A., R. Hornero, A. Mayo, J. Poza, F. Maestú,
project VA108A06 from ‘‘Consejerı́a de Educación de and T. Ortiz. Quantitative magnetoencephalography of
Castilla y León’’ and by the ‘‘Ministerio de Educación spontaneous brain activity in Alzheimer disease: an
exhaustive frequency analysis. Alzheimer Dis. Assoc. Dis.
y Ciencia’’ and FEDER grant MTM2005-08519-C02-
20:153–159, 2006.
01. The authors would like to thank the ‘‘Asociación 15
Fernández, A., F. Maestú, C. Amo, P. Gil, T. Fehr, C.
de Enfermos de Alzheimer’’ (AFAL) for supplying the Wienbruch, B. Rockstroh, T. Elbert, and T. Ortiz. Focal
patients who have participated in this study. temporoparietal slow activity in Alzheimer’s disease
revealed by magnetoencephalography. Biol. Psychiat.
52:764–770, 2002.
16
REFERENCES Fernández, A., A. Turrero, P. Zuluaga, P. Gil, F. Maestú,
P. Campo, and T. Ortiz. Magnetoencephalographic parie-
tal delta dipole density in mild cognitive impairment: pre-
1
Abásolo, D., R. Hornero, P. Espino, D. Álvarez, and J. liminary results of a method to estimate the risk of
Poza. Entropy analysis of the EEG background activity in developing Alzheimer disease. Arch. Neurol. 63:427–430,
Alzheimer’s disease patients. Physiol. Meas. 27:241–253, 2006.
17
2006. Folstein, M. F., S. E. Folstein, and P. R. McHugh. Mini-
2
Aviyente, S., L. A. W. Brakel, R. K. Kushwaha, M. mental state. A practical method for grading the cognitive
Snodgrass, H. Shevrin, and W. J. Willians. Characteriza- state of patients for the clinician. J. Psychiatr. Res. 12:189–
tion of event related potentials using information theoretic 198, 1975.
18
distances measures. IEEE Trans. Biomed. Eng. 51:737–743, Franciotti, R., D. Iacono, S. Della Penna, V. Pizzella, K.
2004. Torquati, M. Onofrj, and G. L. Romani. Cortical rhythms
3
Baraniuk, R. G., P. Flandrin, A. J. E. M. Janssen, and O. reactivity in AD, LBD and normal subjects. A quantitative
J. J. Michel. Measuring time-frequency information con- MEG study. Neurobiol. Aging 27:1100–1109, 2006.
19
tent using the Rényi entropies. IEEE Trans. Inform. Theory Gómez, C., R. Hornero, D. Abásolo, A. Fernández, and
47:1391–1409, 2001. M. López. Complexity analysis of the magnetoencephalo-
4
Berendse, H. W., J. P. A. Verbunt, Ph. Scheltens, B. W. gram background activity in Alzheimer’s disease patients.
van Dijk, and E. J. Jonkman. Magnetoencephalographic Med. Eng. Phys. 28:851–859, 2006.
20
analysis of cortical activity in Alzheimer’s disease: a pilot Hari, R. Magnetoencephalography in clinical neurophysi-
study. Clin. Neurophysiol. 111:604–612, 2000. ological assessment of human cortical functions. In: Elec-
5
Berezianos, A., S. Tong, and N. Thakor. Time-dependent troencephalography: Basic Principles, Clinical
entropy estimation of EEG rhythm changes following Applications, and Related Fields, 5th ed., edited by E.
brain ischemia. Ann. Biomed. Eng. 31:221–232, 2003. Niedermeyer and F. Lopes da Silva. Philadelphia: Lippi-
6
Bird, T. D. Alzheimer’s disease and other primary ncontt Williams & Wilkins, 2005, pp. 1165–1197.
21
dementias. In: Harrison’s Principles of Internal Medicine, Hämäläinen, M., R. Hari, R. J. Ilmoniemi, J. Knuutila,
edited by E. Braunwald, A. S. Fauci, D. L. Kasper, S. L. and O. V. Lounasmaa. Magnetoencephalography—theory,
Hauser, D. L. Longo, and J. L. Jameson. New York: instrumentation, and applications to non-invasive studies
McGraw-Hill, 2001, pp. 2391–2399. of the working human brain. Rev. Mod. Phys. 65:413–497,
7
Blennow, K., M. J. de Leon, and H. Zetterberg. Alzhei- 1993.
22
mer’s disease. Lancet 368:387–403, 2006. Inouye, T., K. Shinosaki, H. Sakamoto, S. Toi, S. Ukai, A.
8
van Cappellen van Walsum, A.-M., Y. A. L. Pijnenburg, Iyama, Y. Katsuda, and M. Hirano. Quantification of EEG
H. W. Berendse, B. W. van Dijk, D. L. Knol, Ph. Schel- irregularity by use of the entropy of the power spectrum.
tens, and C. J. Stam. A neural complexity measure applied Electroencephalogr. Clin. Neurophysiol. 79:204–210, 1991.
23
to MEG data in Alzheimer’s disease. Clin. Neurophysiol. Jeong, J. EEG dynamics in patients with Alzheimer’s dis-
114:1034–1040, 2003. ease. Clin. Neurophysiol. 115:1490–1505, 2004.
152 POZA et al.
24 34
Kannathal, N., M. L. Choob, U. R. Acharyab, and P. K. Rossor, M. Alzheimer’s disease. In: Brain’s Diseases of the
Sadasivana. Entropies for detection of epilepsy in EEG. Nervous System, edited by M. Donaghy. Oxford: Uni-
Comput. Meth. Prog. Biomed. 80:187–194, 2005. versity Press, 2001, pp. 750–754.
25 35
Knopman, D. S., S. T. DeKosky, J. L. Cummings, H. Chui, Rényi, A. Probability Theory. Amsterdam: North-Hol-
J. Corey-Bloom, N. Relkin, G. W. Small, B. Miller, and J. land, 1970.
36
C. Stevens. Practice parameter: diagnosis of dementia (an Sleigh, J. W., D. A. Steyn-Ross, C. Grant, and G. Lud-
evidence-based review). Report of the quality standards brook. Cortical entropy changes with general anaesthesia:
subcommittee of the American Academy of Neurology. theory and experiment. Physiol. Meas. 25:921–934, 2004.
37
Neurology 56:1143–1153, 2001. Stam, C. J. Nonlinear dynamical analysis of EEG and
26
de Leon, M. J., and W. Klunk. Biomarkers for the early MEG: review of an emerging field. Clin. Neurophysiol.
diagnosis of Alzheimer’s disease. Lancet Neurol. 5:198–199, 116:2266–2301, 2005.
38
2006. Stam, C. J., A. M. van Cappellen van Walsum, Y. A. L.
27
McKhann, G., D. Drachman, M. Folstein, R. Katzman, D. Pijnenburg, H. W. Berendse, J. C. de Munck, Ph. Scheltens,
Price, and E. M. Stadlan. Clinical diagnosis of Alzheimer’s and B. W. van Dijk. Generalized synchronization of MEG
disease: report of NINCDS-ADRDA work group under recordings in Alzheimer’s disease: evidence for involvement
the auspices of department of health and human services of the gamma band. J. Clin. Neurophysiol. 19:562–574,
task force on Alzheimer’s disease. Neurology 34:939–944, 2002.
39
1984. Stam, C. J., B. F. Jones, I. Manshanden, A. M. van Cap-
28
Osipova, D., J. Ahveninen, O. Jensen, A. Ylikoski, and E. pellen van Walsum, T. Montez, J. P. Verbunt, J. C. de
Pekkonen. Altered generation of spontaneous oscillations Munck, B. W. van Dijk, H. W. Berendse, and P. Scheltens.
in Alzheimer’s disease. Neuroimage 27:835–841, 2005. Magnetoencephalographic evaluation of resting-state
29
Osipova, D., K. Rantanen, J. Ahveninen, R. Ylikoski, O. functional connectivity in Alzheimer’s disease. Neuroimage
Happola, T. Strandberg, and E. Pekkonen. Source esti- 32:1335–1344, 2006.
40
mation of spontaneous MEG oscillations in mild cognitive Tong, S., A. Berezianos, J. Paul, Y. Zhu, and N. Thakor.
impairment. Neurosci. Lett. 405:57–61, 2006. Nonextensive entropy measure of EEG following brain
30
Poza, J., R. Hornero, D. Abásolo, A. Fernández, and J. injury from cardiac arrest. Physica A 305:619–628, 2002.
41
Escudero. Analysis of spontaneous MEG activity in pa- Tong, S., A. Bezerianos, A. Malhotra, Y. Zhu, and N.
tients with Alzheimer’s disease using spectral entropies. In: Thakor. Parameterized entropy analysis of EEG following
Proc. of the 29th Ann. Int. Conf. of the IEEE EMBS, hypoxic–ischemic brain injury. Phys. Lett. A 314:354–361,
Lyon, France, 2007, pp 6179–6182. 2003.
31 42
Poza, J., R. Hornero, D. Abásolo, A. Fernández, and M. Tsallis, C. Possible generalization of Boltzmann-Gibbs
Garcı́a. Extraction of spectral based measures from MEG statistics. J. Stat. Phys. 52:479–487, 1988.
43
background oscillations in Alzheimer’s disease. Med. Eng. Tsallis, C. Generalized entropy-based criterion for consis-
Phys., 2006 (in press). Available: doi: 10.1016/ tent testing. Phys. Rev. E 58:1442–1445, 1998.
44
j.medengphy.2006.11.006. Vellas, B., S. Andrieu, C. Sampaio, and G. Wilcock. (for
32
Reisberg, B. Functional assessment staging (FAST). Psy- the European Task Force group). Disease-modifying trials
chopharmacol. Bull. 24:653–659, 1988. in Alzheimer’s disease: a European task force consensus.
33
Rosso, O. A., M. T. Martin, A. Figliola, K. Keller, and A. Lancet Neurol. 6:56–62, 2007.
Plastino. EEG analysis using wavelet-based information
tools. J. Neurosci. Methods 153:163–182, 2006.

You might also like