32nd Annual International Conference of the IEEE EMBS
Buenos Aires, Argentina, August 31 - September 4, 2010
Abstract— The aim of this study was to explore the ability of
several spectral entropies and disequilibrium measures to
discriminate between spontaneous magnetoencephalographic
(MEG) oscillations from 18 mild cognitive impairment (MCI)
patients and 24 controls. The Shannon spectral entropy (SSE),
Tsallis spectral entropy (TSE), and Rényi spectral entropy
(RSE) were calculated from the normalized power spectral
density to evaluate the irregularity patterns. In addition, the
Euclidean (ED) and Wootters (WD) distances were computed
as disequilibrium measures. Results revealed statistically
significant lower SSE and TSE(2) values for MCI patients than
for controls (p < 0.05) in the right lateral region of the brain.
ED also obtained statistically significant lower values for MCI
patients than for controls using the (p < 0.05) in the right
lateral region of the brain. These findings suggest that MCI is
associated with a significant decrease in irregularity of MEG
activity. In addition, the highest accuracy of 64.3% was
achieved by the SSE. We conclude that measures from
information theory can be useful to both characterize abnormal
brain dynamics and help in MCI detection.
ILD cognitive impairment (MCI) is a disorder of the
brain, which is considered an intermediate clinical
stage between the cognitive decline of normal aging and the
more serious problems caused by Alzheimer’s disease (AD)
. Although MCI can affect multiple brain areas, such as
those involved in though and action, the most common
variety of MCI causes memory problems. Thus, MCI
represents a subtle but measurable memory disorder, in
which patients experience memory problems greater than
normally expected with aging, but do not show other
symptoms of dementia, such as impaired judgment or
Likewise AD, MCI is a heterogeneous disease, where
several subtypes can be distinguished . Due to this fact, it
is complex to determine whether a potential MCI patient is
exhibiting cognitive changes of normal aging or the earliest
stages of dementia . Nowadays, there is not an effective
test to detect MCI. It is diagnosed by excluding other
Manuscript received March 30, 2010. This work was supported in part
by the project TEC2008-02241 from “Ministerio de Ciencia e Innovación”.
Asterisk indicates corresponding author.
R. Bruña, J. Poza*, R. Hornero, J. Escudero, and C. Gómez are with the
Biomedical Engineering Group (GIB), Dpt. TSCIT, University of
Valladolid, Camino del Cementerio s/n, 47011-Valladolid, Spain (phone:
+34 983 423000, ext. 5569; fax: +34 983 423667; e-mail:
A. Fernández is with the Centro de Magnetoencefalografía Dr. Pérez-
Modrego, Complutense University of Madrid, 28040-Madrid, Spain.
conditions that might be causing the signs and symptoms.
An exhaustive examination is carried out, including a
physical exam, a neurological test, a mental status exam,
laboratory tests and brain scans .
Despite the difficulties to achieve a diagnosis, an early
and accurate identification of MCI should be attempted. A
previous diagnosis of MCI is considered an important risk
factor for the development of AD. Most patients with MCI
develop a progressive decline in their thinking abilities over
time, and AD is usually the underlying cause . On the
other hand, an early diagnosis is important in the case of
pharmacologic interventions. Though there is no cure for
MCI, current treatments may help to reduce cognitive
problems. Moreover, nonpharmacologic strategies could be
applied since the appearance of the first symptoms. As a
conclusion, new tools are needed to help in MCI detection
Given the fact that MCI affects the brain cortex,
magnetoencephalographic (MEG) signals can reflect
anatomical and functional deficits of the brain. EEG and
MEG recordings are related . However, scalp EEG is
much more affected by the conductivities of the skull and
the scalp than MEG . Due to this fact, it can be
hypothesized that MEG might provide a more accurate
vision of the cortical activity function than scalp EEG .
The alterations that MCI produces in electromagnetic
brain oscillations have been analyzed by means of EEG –
, whereas only a few studies have focused on MEG .
In this sense, several differences have been found in the
MEG activity between MCI and AD patients. MEG studies
have shown increased slow rhythms and reduced fast
activity in AD compared to MCI, using median frequency
and alpha frequency , . Some authors have also
reported an increase of complexity in MCI patients’ MEG
background activity using non-linear measures . In
addition, studies analyzing functional connectivity have
found different patterns of coherence and synchronization
likelihood . Nevertheless, only subtle differences have
been observed between MCI patients and elderly healthy
This study is a first approach to explore the ability of
several measures from information theory to characterize
MEG rhythms in MCI. The proposed measures are based on
the Fourier transform (FT). Firstly, the FT was used to
compute the power spectral density (PSD) for each MEG
recording in five brain regions. Shannon spectral entropy
Analysis of Spontaneous MEG Activity in Mild Cognitive
Impairment Using Spectral Entropies and Disequilibrium Measures
Ricardo Bruña, Jesús Poza*, Member, IEEE, Carlos Gómez, Student Member, IEEE, Alberto
Fernández, and Roberto Hornero, Member, IEEE
978-1-4244-4124-2/10/$25.00 ©2010 IEEE6296
(SSE), Tsallis spectral entropy (TSE), and Rényi spectral
entropy (RSE) were then calculated to explore the
irregularity patterns in terms of the flatness of the spectrum.
In addition, the Euclidean (ED) and the Wootters (WD)
distances were employed to provide an alternative way to
quantify the irregularity of MEGs by means of
II. MATERIALS AND METHODS
A. Subjects and MEG Recording
Eighteen patients (8 men and 10 women, age = 75.4 ± 5.5
years, mean ± standard deviation SD) from the “Asociación
de Familiares de Enfermos de Alzheimer” and twenty-four
cognitively normal volunteers (9 men and 15 women, age =
71.7 ± 6.5 years, mean ± SD) participated in the study. All
patients were diagnosed following Petersen’s criteria .
Their cognitive function was evaluated using the Mini-
Mental State Examination (MMSE). MCI patients obtained a
mean MMSE score of 25.7 ± 1.8 (mean ± SD), whereas
controls had a mean MMSE score of 28.9 ± 1.2 points. No
significant differences were observed in the mean age of
both groups (p > 0.05, Mann-Whitney U test). MCI patients
and controls were not taking any medication that could
affect the central nervous system. Informed consent was
obtained from all controls and patients’ caregivers. The
study was approved by the local ethics committee.
Five minutes of spontaneous MEG activity were recorded
for each subject using a 148-channel whole-head
magnetometer (MAGNES 2500 WH, 4D Neuroimaging),
placed in a magnetically shielded room in the “Centro de
Magnetoencefalografía Dr. Pérez Modrego” of the
Complutense University of Madrid. MEGs were acquired at
a sampling rate of 678.17 Hz with subjects in a relaxed state,
awake and with eyes closed. A 0.1-200 Hz hardware band-
pass filter and a 50 Hz notch filter were applied. Each
recording was processed with a low-pass filter before
downsampling by a factor of 4 to reduce the data length.
Artifact-free epochs of length 10 s (28.4 ± 3.6 artifact-free
epochs per channel and subject, mean ± SD) were selected
for further analysis. In addition, outer sensors were excluded
from the analysis due to the low signal-to-noise ratio. Prior
to time-frequency analysis, each MEG signal of 1696
samples was digitally band-pass filtered between 1 and 70
B. Definition of Spectral Parameters
The MEG irregularity can be indirectly analyzed applying
several measures to their power-frequency distributions .
In the present work, five spectral parameters based on the FT
were calculated to quantify the irregularity of the signal:
SSE, TSE, RSE, ED and WD. MEG recordings were filtered
between the cut-off frequencies, f1 = 1 Hz and f2 = 70 Hz,
and segmented. Each segment contained 10 s of non-
overlapping data (1696 samples). The PSD for each MEG
segment was calculated as the FT of the autocorrelation
function. PSDs were averaged to obtain the mean power
spectrum in five brain regions. Finally, the PSD was
normalized to scale from 0 to 1 to obtain the normalized
Three entropies were calculated from the PSDn, which
was considered as a probability distribution. The SSE is a
disorder quantifier, which has been previously employed to
characterize MEG irregularity in AD , . Its
definition is based on the Shannon’s entropy and quantifies
the flatness of the power spectrum .
The TSE is a generalized information measure, which
extends the notion of the SSE. It is a non-logarithmic
entropy, what makes it useful to explore the properties from
a new mathematical framework. The TSE is controlled by
the entropic index, q ∈ ℜ, which can be considered as a
measure of the degree of non-extensivity . Hence, it is
possible to obtain the SSE from the TSE in the limit q → 1.
Its definition is given by,
The RSE is an extensive generalized information measure,
which can be reduced to the SSE entropy in the limit q → 1
. This measure can be used to quantify the uncertainty of
a signal, and its definition is given by,
In addition to the entropies, two disequilibrium measures
were calculated as estimators of the irregularity in a signal.
The original measure of disequilibrium is the ED, which is
defined as the distance in the probability space between the
considered distribution and the uniform distribution . It
can be then considered as a measure of irregularity, and it is
defined as follows,
Another definition of disequilibrium is the WD. It
provides an alternative framework to quantify the distance
between distributions . Its definition is given by,
All calculations were carried out with the software
PSDf PSDq TSE
nf PSDq RSE
Nf PSD ED
package Matlab (version 7.0; Mathworks, Natick, MA).
C. Statistical Analysis
The Kolmogorov–Smirnov test was used to assess the
normal distribution of the variables, while homocedasticity
was evaluated with Levene test. After the descriptive
analysis, we found that the log-transformed values met
parametric test assumptions. Therefore, two-way ANOVAs
(with group as between-subject factor and brain region as
within-subject factor) with contrast were performed to
explore statistical significance for each parameter (α = 0.05)
at five brain regions (anterior, central, left lateral, posterior,
and right lateral).
Receiver operating characteristics (ROC) curves with a
leave-one-out cross-validation (LOO-CV) procedure were
used to visually evaluate the ability of each parameter to
distinguish between MCI patients and controls. In order to
quantify the parameter’s performance, we used the area
under ROC curve (AUC) and the accuracy obtained with the
All statistical analyses were performed using SPSS
software (version 15.0; SPSS Inc, Chicago, IL).
III. RESULTS AND DISCUSSION
The PSDn of the 10 s epochs was computed for each
channel. Results were averaged over five brain regions for
each subject (anterior, central, left lateral, posterior, and
right lateral), which were defined according to previous
works , . Spectral entropies and disequilibrium
measures were then computed. In the case of TSE and RSE
the entropic index was set to 2 and 3.5 respectively,
according to the optimal values obtained in a previous
Table I summarizes the mean values and the standard
deviations of the parameters for each group that obtained
significant p-values, together with the results of the
statistical analysis. MCI patients obtained significantly lower
SSE and TSE(2) values (p < 0.05) than controls in the right
lateral region, which suggests a significant decrease in the
irregularity of MCI patients’ MEGs. On the other hand, MCI
patients exhibited significantly higher ED values (p < 0.05)
than controls in the right lateral region. This issue indicates
that MEG recordings from MCI patients have a PSDn more
different to the equiprobable distribution than healthy
controls, which suggest a decrease in irregularity again. The
results are in agreement with the decrease in irregularity
reported in previous studies which analyzed spontaneous
MEG activity in AD using spectral entropies ,  and
non-linear parameters , . Nevertheless, to the best of
our knowledge this is the first work addressing the
characterization of the irregularity patterns in MCI. Related
to the previous issue, Fernández et al.  observed a trend
to decrease in non-linear complexity of MCI patients, though
they did not obtained significant differences.
Fig. 1 depicts the ROC curves for the significant
parameters, previously presented in Table I. A summary of
their characteristics can be observed in Table II, where
accuracy and AUC applying LOO-CV are shown. The SSE
achieved both the highest AUC of 0.664 ± 0.014 and
accuracy of 64.3%. It is noteworthy that previous studies
have suggested the usefulness of SSE to discriminate
between AD patients and controls using MEG recordings
, , . The TSE(2) and the ED obtained slightly
lower classification statistics than the SSE, with an AUC of
0.655 ± 0.014 and an accuracy of 61.9%. Studies analyzing
functional connectivity and lentification of the oscillatory
cerebral activity in MCI and elderly healthy subjects
obtained similar AUC and accuracy statistics .
Nevertheless, they did not apply a LOO-CV procedure. In
addition, an advantage of the methods used in the present
study, compared to previous efforts, is their ability to carry
out a analysis of the underlying dynamics, at the same time
an easily interpretation of the results can be achieved.
Therefore, it is possible to obtain an alternative description
of the brain dynamics in MCI to conventional spectral and
Finally, some limitations should be addressed in future
works. We used a small sample size. Likewise, the number
of subjects enrolled in the study should be increased. The
analysis should also be extended to other neurodegenerative
disorders with similar alterations to those observed in MCI
and AD. In addition, as a first approach to the problem, we
set the entropic values for TSE and RSE taking into account
the results obtained in a previous study, where AD patients
and controls were compared. Further works should be
performed to accurately explore the influence of the index q
This preliminary study was performed to analyze the
spontaneous MEG activity in MCI patients and controls by
means of three spectral entropies (SSE, TSE, and RSE), and
two disequilibrium measures (ED and WD). A significant
loss in irregularity was found in MCI patients’ MEGs.
Furthermore, our results suggest that the entropies and the
disequilibrium measures based on PSD could be useful to
both characterize the abnormal brain dynamics associated
with MCI and help in diagnosis. The measures from
information theory can be used to yield new information
about MEG rhythms in MCI in comparison with that
obtained using conventional spectral and non-linear
AVERAGE VALUES (MEAN ± SD) FOR SIGNIFICANT PARAMETERS IN BOTH
GROUPS AND THE RESULTS OF THE STATISTICAL ANALYSIS
Parameter Region Controls MCI patients p-value
Right Lateral 0.8750 ± 0.0253 0.8566 ± 0.0281
Right Lateral 0.9985 ± 0.0005 0.9979 ± 0.0010
Right Lateral 0.0015 ± 0.0005 0.0021 ± 0.0010
( ) 2
REFERENCES Download full-text
 R. C. Petersen and S. Negash, “Mild cognitive impairment: an
overview,” CNS Spectr., vol. 13, pp. 45–53, Jan. 2008.
 R. C. Petersen, R. O. Roberts, D. S. Knopman, B. F. Boeve, Y. E.
Geda, R. J. Ivnik, G. E. Smith, and C. R. Jack Jr., “Mild cognitive
impairment: ten years later,” Arch.Neurol., vol. 66, pp. 1447–1455,
 F. Portet, P. J. Ousset, P. J. Visser, G. B. Frisoni, F. Nobili, P.
Scheltens, B. Vellas J. Touchon, and MCI Working Group of the
European Consortium on Alzheimer’s Disease (EADC), “Mild
cognitive impairment (MCI) in medical practice: A critical review of
the concept and new diagnostic procedure. Report of the MCI
Working Group of the European Consortium on Alzheimer’s
Disease,” J. Neurol. Neurosurg. Psychiatry, vol. 77, pp. 714–718, Jun.
 R. C. Petersen, “Early diagnosis of Alzheimer’s disease: is MCI too
late?,” Curr. Alzheimer Res., vol. 6, pp. 324–330, Aug. 2009.
 S. Rampp and H. Stefan, “On the opposition of EEG and MEG,” Clin.
Neurophysiol., vol. 118, pp. 1658−1659, Aug. 2007.
 R. Hari, “Magnetoencephalography in clinical neurophysiological
assessment of human cortical functions,” in Electroencephalography:
Basic Principles, Clinical Applications, and Related Fields, 5th ed., E.
Niedermeyer and F. Lopes da Silva, Eds. Philadelphia: Lippincontt
Williams & Wilkins, 2005, pp. 1165–1197.
 P. L. Nunez, B. M. Wingeier, and R. B. Silberstein, “Spatial-Temporal
structures of human alpha rhythms: Microcurrents sources, multiscale
elements, and global binding of local networks,” Hum. Brain Mapp.,
vol. 13, pp. 125–164, Jul. 2001.
 D. V. Moretti, G. B. Frisoni, M. Pievani, S. Rosini, C. Geroldi, G.
Binetti, and P.M. Rossini, “Cerebrovascular disease and hippocampal
atrophy are differently linked to functional coupling of brain areas: An
EEG coherence study in MCI subjects,” J. Alzheimers Dis., vol. 14,
pp. 285–299, Jul. 2008.
 P. M. Rossini, C. del Percio, P. Pasqualetti, E. Cassetta, G. Binetti, G.
Dal Forno, F. Ferreri, G. Frisoni, P. Chiovenda, C. Miniussi, L. Parisi,
M. Tombini, F.Vecchio, and C. Babiloni, “Conversion from mild
cognitive impairment to Alzheimer’s disease is predicted by sources
and coherence of brain
Neuroscience, vol. 143, pp. 793–803, Oct. 2006.
 C. J. Stam,Y.Van DerMade,Y. A. L. Pijnenburg, and Ph. Scheltens,
“EEG synchronization in mild cognitive impairment and Alzheimer’s
disease,” Acta Neurol. Scand., vol. 108, pp. 90–96, Aug. 2003.
 C. J. Stam, “Use of magnetoencephalography (MEG) to study
functional brain networks in neurodegenerative disorders,” J. Neurol.
Sci., vol. 289, pp. 128–134, Feb. 2010.
 D. Osipova, K. Rantanen, J. Ahveninen, R. Ylikoski, O. Paola, T.
Strandberg, and E. Pekkonen, “Source estimation of spontaneous
MEG oscillations in mild cognitive impairment,” Neurosci. Lett., vol.
405, pp. 57–61, Sep. 2006.
 A. Fernández, R. Hornero, A. Mayo, J. Poza, P. Gil-Gregorio, and T.
Ortiz, “MEG spectral profile in Alzheimer’s disease and mild
cognitive impairment,” Clin. Neurophysiol., vol. 117, pp. 306–314,
 A. Fernández, R. Hornero, C. Gómez, A. Turrero, P. Gil-Gregorio, J.
Matías-Santos, and T. Ortiz, “Complexity analysis of spontaneous
brain activity in Alzheimer’s Disease and Mild Cognitive
Impairment,” Alzheimer Dis. Assoc. Disord., to be published.
 C. Gómez, C. J. Stam, R. Hornero, A. Fernández, and F. Maestú,
“Disturbed beta band functional connectivity in patients with mild
cognitive impairment: An MEG study,” IEEE Trans. Biomed. Eng.,
vol. 56, pp. 1683–1690, Jun. 2009.
 R. C. Petersen, R. Doody, A. Kurz, R. C.Mohs, J. C.Morris, P. V.
Rabins, K. Ritchie, M. Rossor, L. Thal, and B.Winblad, “Current
concepts in mild cognitive impairment,” Arch. Neurol., vol. 58, pp.
1985–1992, Dec. 2001.
 J. Poza, J. Escudero, R. Hornero, A. Fernández, and C. I. Sanchez,
“Regional analysis of spontaneous MEG rhythms in patients with
Alzheimer’s disease using spectral entropies,” Ann. Biomed. Eng., vol.
36, pp. 141–152, Jan. 2008.
 J. Poza, R. Hornero, D. Abásolo, A. Fernández, and M. García,
“Extraction of spectral based measures from MEG background
oscillations in Alzheimer’s disease,” Med. Eng. Phys., vol. 29, pp.
1073–1083, Dec. 2007.
 T. Inouye, K. Shinosaki, H. Sakamoto, S. Toi, S. Ukai, A. Iyama, Y.
Katsuda, and M. Hirano, “Quantification of EEG irregularity by use of
the entropy of the power spectrum,” Electroencephalogr. Clin.
Neurophysiol., vol. 79, pp. 204–210, Sep. 1991.
 C. Tsallis, “Possible generalization of Boltzmann-Gibbs statistics,” J.
Stat. Phys., vol. 52, pp. 479–487, Jul. 1988.
 A. Rényi, Probability theory, Amsterdam: North-Holland, 1970.
 R. López-Ruiz, H. L. Mancini, and X. Calbet, “A statistical measure
of complexity”, Phys. Lett. A., vol. 209, 321–326, Dec. 1995.
 W. K. Wootters, “Statistical distance and Hilbert space,” Phys. Rev.
D, vol. 23, pp. 357–362, Jan. 1981.
 R. Hornero, J. Escudero, A. Fernández, J. Poza, and C. Gómez,
“Spectral and Non-linear Analyses of MEG Background Activity in
Patients with Alzheimer’s Disease,” IEEE Trans. Biomed. Eng., vol.
55, pp. 1658–1665, Jun. 2008.
 R. Hornero, D. Abásolo, J. Escudero, and C. Gómez, “Nonlinear
analysis of electroencephalogram and magnetoencephalogram
recordings in patients with Alzheimer's disease,” Philos. Transact. A
Math. Phys. Eng. Sci., vol. 367, pp. 317–336, Jan. 2009.
ACCURACY AND AUC (MEAN ± SD) OF THE ROC ANALYSIS WITH A
LOO-CV PROCEDURE USING THE SIGNIFICANT PARAMETERS
Parameter Region Accuracy (%) AUC
Right Lateral 64.3 0.664 ± 0.014
0.655 ± 0.014
0.655 ± 0.014
( ) 2
Right Lateral 61.9
ED Right Lateral 61.9
Fig. 1. ROC curves for the significant parameters showing the
discrimination between MCI patients and controls.