Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey
Autism spectrum disorders (ASD) are pervasive neurodevelopmental conditions characterized by impairments in reciprocal social interactions, communication skills, and stereotyped behavior. Since EEG recording and analysis is one of the fundamental tools in diagnosis and identifying disorders in ne...
Збережено в:
| Опубліковано в: : | Нейрофизиология |
|---|---|
| Дата: | 2014 |
| Автори: | , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
Інститут фізіології ім. О.О. Богомольця НАН України
2014
|
| Теми: | |
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/148271 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey / M. Hashemian, H. Pourghassem // Нейрофизиология. — 2014. — Т. 46, № 2. — С. 197-209. — Бібліогр.: 76 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| _version_ | 1859627055012904960 |
|---|---|
| author | Hashemian, M. Pourghassem, H. |
| author_facet | Hashemian, M. Pourghassem, H. |
| citation_txt | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey / M. Hashemian, H. Pourghassem // Нейрофизиология. — 2014. — Т. 46, № 2. — С. 197-209. — Бібліогр.: 76 назв. — англ. |
| collection | DSpace DC |
| container_title | Нейрофизиология |
| description | Autism spectrum disorders (ASD) are pervasive neurodevelopmental conditions characterized
by impairments in reciprocal social interactions, communication skills, and stereotyped
behavior. Since EEG recording and analysis is one of the fundamental tools in diagnosis
and identifying disorders in neurophysiology, researchers strive to use the EEG signals for
diagnosing of individuals with ASD. We found that studies on the ASD diagnosis using EEG
techniques could be divided into two groups, where analysis was based on either comparison
techniques or pattern recognition techniques. In this paper, we try to explain these two sets of
algorithms along with their applied methods and results. Ultimately, evaluation measures of
diagnosis algorithms are discussed
Розлади аутистичного спектра (autism spectrum disorders –
ASD) – це глибокі відхилення розвитку нервової
сфери, що характеризуються порушенням соціальних
взаємодій, комунікативних навичок та стереотипної
поведінки. Оскільки реєстрація та аналіз ЕЕГ є одними
із фундаментальних засобів діагностики та ідентифікації
нейрофізіологічних розладів, дослідники намагаються
використовувати ЕЕГ-сигнали для діагностики ASD у тих
або інших осіб. Як ми встановили, дослідження, спрямовані
на діагностику ASD із застосуванням ЕЕГ-методик, можуть
бути поділені на дві групи, коли аналіз базується або на
техніці порівнянь, або на техніці розпізнавання образів.
У цьому огляді ми намагались описати застосування двох
відповідних комплексів алгоритмів, а також методики
їх використання та отримані результати. Нарешті,
обговорюється порівняльна ефективність вказаних
алгоритмів діагностування.
|
| first_indexed | 2025-11-29T12:08:46Z |
| format | Article |
| fulltext |
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 197
UDC 616.89+57.02+616-071
M. HASHEMIAN¹ and H. POURGHASSEM1
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS:
A SURVEY
Received May 29, 2013
Autism spectrum disorders (ASD) are pervasive neurodevelopmental conditions characterized
by impairments in reciprocal social interactions, communication skills, and stereotyped
behavior. Since EEG recording and analysis is one of the fundamental tools in diagnosis
and identifying disorders in neurophysiology, researchers strive to use the EEG signals for
diagnosing of individuals with ASD. We found that studies on the ASD diagnosis using EEG
techniques could be divided into two groups, where analysis was based on either comparison
techniques or pattern recognition techniques. In this paper, we try to explain these two sets of
algorithms along with their applied methods and results. Ultimately, evaluation measures of
diagnosis algorithms are discussed.
Keywords: autism spectrum disorders (ASD), EEG, feature extraction, classification,
ASD diagnosis algorithms.
¹ Department of Electrical Engineering, Najafabad Branch, Islamic Azad
University, Isfahan, Iran.
Correspondence should be addressed to H. Pourghassem
(e-mail: h_pourghasem@iaun.ac.ir).
INTRODUCTION
Autism spectrum disorders (ASD) are pervasive
neurodevelopmental conditions characterized
by impairments in reciprocal social interactions,
communication skills, and stereotyped behavior [1].
The ASD are composed of five disorders: autism,
pervasive development disorder-not otherwise
specified (PDD-NOS), Asperger’s syndrome (AS),
childhood disintegrative disorder (CDD), and Rett’s
disorder (RD) [2, 3]. Different nations have emphasis
to the studies of ASD differently. For example,
developing countries have paid further less attention to
the subject matter. Since, the major part of performed
studies on ASD prevalence has been reported in North
America and European countries [4]. According to a
few scientific investigations, there has been a growing
rate of ASDs in recent years. For example, it has
been estimated that, in the United Kingdom, 157 of
10,000 primary school children have, on average,
ASD [5]. Also, the prevalence rate of ASD among
the British adults has been estimated as 98 of 10,000
[6]. Likewise, in a research performed in the United
States, the prevalence rate of ASD among 8-year-old
children was estimated as 90 of 10,000 [7]. During
1966 to 2008, the prevalence rates of autism, PDD-
NOS, and AS were estimated 20, 30, and 2 of 10,000,
respectively [8]. Based on a preliminary research, the
prevalence rates for Iranian children were found to
be 19 and 5 of 1,000 for autistic and AS disorders,
respectively [9]. Moreover, in another study, the
number of Iranian university students with ASD was
revealed to be 120 of 1,000. Additionally, the number
of males (as compared with that of females) was
significantly higher [10].
The diagnosis of ASD is not an easy process and
generally requires estimation of certain behavioral and
cognitive characteristics [11]. Today, the researchers
are trying to find the ASD diagnostic approaches
through electrophysiological and neuroimaging
techniques. Since EEG recording and analysis is one
of fundamental tools in diagnosis and identifying
disorders in neurophysiology, the researchers strive to
use the EEG technique for diagnosing of individuals
with ASD. A group of the studies have investigated
that EEG signals of individuals with ASD are relevant
to age- and intelligence quotient (IQ)-matched control
subjects based on different conditions. In these
studies, comparative methods and statistical criteria to
analyze the results have been used. In became possible
to identify some characteristics of the brain signals
that explicitly differentiate between the EEG signals
of normal individuals and individuals with ASD.
However, it is noteworthy to mention that these studies
ОБЗОРЫ
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2198
M. HASHEMIAN and H. POURGHASSEM
have only sometimes provided identical results. The
next group of studies has taken large steps in the path
of diagnosing ASD by using pattern recognition and
classification techniques that they have been able to
separate the brain signal patterns of normal individuals
from those affected with ASD. In other words, the
consequences of studies provide high-performance
diagnostic algorithms. Therefore, we found that the
researches on brain signal processing of individuals
with ASD could be divided into two groups where
analysis was based either on comparison techniques or
on pattern recognition techniques (Fig. 1). These two
groups of studies represent developmental progress
over the decade in the ASD detection.
In this paper, we try to present an overview of the
recent researches on EEG-based diagnosis approached
of ASD. Moreover, this overview provides a rather
comprehensive outlook on the ASD. We divided the
performed studies in ASD analysis into the two above-
mentioned main groups. In each group, we describe the
principle of presented algorithms, evaluation measures
of ASD detection and diagnosis, and reported results.
Moreover, we mention the strong and weak points of
the algorithms presented in the literature.
ASD ANALYSIS BASED ON EEG
COMPARISON TECHNIQUES
In the studies of ASD diagnosis based on comparison
techniques, a comparison between the EEG signals
of normal individuals to those affected with ASD in
terms of their age and IQ is carried out. Two main
factors are considered in the process of EEG signal
comparison. Firstly, what types of characteristics are
being examined? In the EEG signals, the valuable
information needing to be extracted and compared has
been concealed. The extracted information is called
“features,” and each one of these features is obtained
by using various signal processing methods and
different scenarios. Secondly, what statistical method
should be used to measure the difference between the
EEG ASD-related and non-ASD samples according to
the characteristics?
In each study, an appropriate statistical method
based on comparison models, examined factors, and
related hypotheses is used for ASD detection. In
majority of these statistical techniques, like analysis
of variance (ANOVA) and the t-test, the result of
compared process is denoted by index P (probability
of the zero hypothesis). If the value of P is less
than the preset significance level (P < alpha), it
represents a statistically significant difference in
the evaluated characteristics of samples. In each
process, the significance level is determined based
on the acceptable error level. In most studies, values
of 0.01 or 0.05 are considered for the alpha value.
When the P value on only slightly greater than, e.g.,
0.05, such a situation can be interpreted as the absence
of a significant difference between the compared
values but as a trend toward such a difference.
EEG SIGNAL FEATURES FOR ASD
ANALYSIS
EEG Rhythms. In ASD analysis, EEG rhythms
are the most common used features based on the
comparison technique. The EEG rhythms according
to their frequency bands are most commonly
divided as follows. The delta rhythm corresponds to
2-4 Hz, theta rhythm, to 4-8 Hz, alpha rhythm and mu
rhythm, to 8-13 Hz, beta rhythm, to 13-30 Hz, and
gamma rhythm, to the frequencies higher than 30 Hz.
In some studies, the proposed borders between the
above ranges can be slightly (insignificantly) diferent.
The EEG rhythms have a crucial role for perceiving
brain functions. For example, it seems that working
memory-related processes are marked as oscilations
in the EEG theta frequencies [12-14]. Concerning
the cognitive processes, three alpha subrhythms
have recently been found, the lower-1 alpha (with
6-8 Hz oscilations) responding to cognitive processes
named as “alertness,” the lower-2 alpha (with 8-10 Hz
fluctuations) that seems to be related with attentional
demands [15], and, finally, upper alpha (with
10-12 Hz oscilations) that seems to be related with
stimulus features and/or semantic processes of
EEG signal of
individuals with ASD
Analysis based on
comparison techniques
Analysis based on pattern
recognition techniques
Identifycation of the EEG
signal characteristics
typical of an individual
with ASD
Diagnostic algorithms
F i g. 1. Categorization of the reported researches.
Р и с. 1. Категоризація досліджень, представлених в огляді.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 199
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
memory [16-19].
The so-called mu rhythm is typically maximum
over the sensorimotor cortex at the resting state and is
attenuated by voluntary movements or somatosensory
stimulation. This rhythm is slightly affected by
visual stimulation. In other words, the mu rhythms
suggest the presence of sensorimotor processing in
frontoparietal networks. At the same time, the classical
alpha rhythms suggest the initial visual processing in
the occipital networks [20].
Desynchronization of the beta rhythm usually occurs
during motor activities. The synchronization, however,
occurs immediately after the movement (beta rhythm
rebound) [21]. These processes represent the action of
the motor cortices [22-25]. The reactions of the beta
rhythm have been also recorded when observing other
movements and during motor imagery. Additionally,
the gamma band is observable at the existence of the
visual and/or auditory motor tasks [26-31].
Absolute Power or Relative Power. The absolute
spectral power (ASP) within a given frequency band
corresponds to the area underneath of spectral curves
for the respective frequency band. The relative spectral
power (RSP) is a percentage value that compared the
absolute power within a given frequency band to the
total (integral) absolute power for the entire frequency
range. In other words, the relative power, or the band
relative intensity ratio (RIR), can be defined for each
frequency band i as
(1)
Coherence. Coherence is a benchmark of coupling
between two different time series in the frequency
domain. The estimated coherence can indicate the
“coupling” of functional association between two
brain regions [32]. The coherence between two EEG
channels presents the linear relationship of these two
channels at a specific frequency. Mathematically, the
coherence is calculated as
(2)
(3)
where X i(f) and X j(f) are the (complex) Fourier
transforms of time series xi (t) and xj (t) of channels i
and j, respectively. Sj (f) is the cross-spectrum function
where operator “*” means complex conjugation,
and means the expectation value. In practice, the
expectation value can only be estimated as an average
over a sufficiently large number of epochs [33]. The
estimated coherence is a value within [0, 1] range. If
the value of coherence function is calculated as zero
(i.e., in the frequency fo, Cij (fo) = 0), it indicates that
the activities of signals in this frequency are linearly
independent. At the same time, a value of one, i.e.,
Cij (fo) = 1, gives the maximum linear correlation for
this frequency [34]. Figure 2 indicates a model of the
applied analysis based on the coherence measure in
two groups of children (ASD and non-ASD) [35].
According to this analysis, the dotted and solid lines,
respectively, show the significant differences for
P < 0.05 and P < 0.01 in terms of the coherence values
for three frequency bands (gamma, alpha, and beta)
between these two groups.
F i g. 2. A model of the applied
analysis based on coherence measure
[35] .
Р и с. 2. Модель прикладного
аналізу, базована на оцінці
когерентності [35].
A B C
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2200
M. HASHEMIAN and H. POURGHASSEM
Mu Wave Suppression. The properties of the
mu frequency band can be used as a technique to
investigate human mirror neuron functioning [36].
At the resting state, the synchronous action of the
neurons in the sensorimotor cortex creates large mu
oscillations. When individuals execute or observe
a movement, the power of these mu oscillations is
attenuated. This phenomenon is called mu wave
suppression [36-39]. The amplitude reduction of
mu oscillations indicates desynchronization of the
underlying neurons, reflecting greater levels of active
processing during motor movement and observation
[38, 39].
Cordance. Cordance is a measurable EEG factor
to determine the cerebral blood flow perfusion and
metabolism. In fact, this factor is an incorporation of
both relative and absolute power measures to produce
characteristics that are more strongly correlated
with local cerebral perfusion than each separate
measure [40]. The cordance is calculated based on an
algorithm consisting of three steps. Firstly, the values
of EEG power are calculated by an arrangement of
reattributional electrodes. The reattributed power is
the average of power values from pairs of electrodes
that share a common electrode. [41]. In the second
step, the values of relative and absolute powers of each
individual EEG recording are statistically standardized
among the electrode sites, engaging a z-transformation
for the each electrode site for corresponding frequency
band f. According to this way, the values of Anorm(s,f)
and Rnorm(s,f) are determined. Finally, the values of
cordance for each electrode site s and its corresponding
frequency band f are determined by the following
relation [42]:
Cordance(s, f) = Anorm (s, f) + Rnorm (s, f)
Multi-Scale Entropy. Multi-scale entropy (MSE) is
a computational method for quantifying the complexity
of a time series by calculating the sample entropy
(SF) over several time scales, with utilizing a coarse-
graining procedure [43, 44]. The SF is a measure of
irregularity of a time-series in an EEG time-series x
= {x1,...xi,...xN}, defined as negative of the logarithmic
conditional probability that two similar sequences of m
consecutive data points will remain similar at the next
point (m + 1) [45-47].
(5)
where Cm (r) = , A is the number of pairs (i, j) with
|xi
m – xj
m|<r, i = j, and B is the number of all probable pairs ,
where |xi
m – xj
m|<r denotes the distance
between vectors xi
m and xj
m with dimension m, r is the
tolerable distance between two vectors (in terms of
the standard deviation fraction of the time-series), and
N is the length of time-series. For MSE analysis, the
EEG time-series x = {x1,...xi,...xN} is coarse-grained
into consecutive time-serie {yj
τ} corresponding to the
scale factor (SF) τ. Firstly, the original time-series is
divided into non-overlapping windows of length τ, and
then the data points inside each window are averaged.
Therefore, each coarse-grained time-series is defined
as
, (7)
Eventually, SE is calculated for each time-series {yj
τ}
[47]. Figure 3 illustrates the diagram of the coarse-
graining procedure [44].
ALGORITHMS BASED ON COMPARISON
TECHNIQUES
The reported studies in the comparison-based
algorithms could differ from each other regarding
several views, including (i) age and IQ of the
participants in each experiment, (ii) conditions under
which the EEG signals have been registered, (iii)
the features that have been extracted and compared
Original
data
Scale 2
x1
x1
x1
x4
x4
y1
y1
y1
x2
x2
x2
x5
x5
y2
y2
y2
x3
x3
x3
x6
x6
xi
xi
xi
xi + 1
xi + 1 xi + 2
y3
y3
yj= xi
Scale 3
yj =
xi + xi+1
2
yj =
xi + xi+1+ xi+2
3
yj
(τ) = ∑
jτ
i = (j−1)τ+1
xi ,
1
τ
N1 ≤ j ≤ τ
F i g. 3. Diagram of the coarse-graining procedure [44].
Р и с. 3. Діаграма процедури огрублення дискретизації [44].
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 201
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
in both groups and in each experiment, and (iv) the
factors that affect the results of every experiment.
We briefly explain below the conditions and factors
examined in each study along with their results for
state-of-the-art studies that have been presented in the
literature. For example, Daoust et al. [48] investigated
EEGs of IQ-matched 9 persons with ASD (ages 12 to
53 years) and 8 control participants (ages 8 to
56 years) recorded under two conditions, REM sleep
and wakefulness. In this study, the power spectral
analysis was performed in four frequency bands:
delta (0.75-3.5 Hz), theta (4.0-7.75 Hz), alpha
(8.0-12.75 Hz), and beta (13.0-19.75 Hz). The report
of these authors [48] indicated that individuals
suffering from ASD, when being in REM sleep
in comparison with the controls, demonstrated a
significantly lower absolute beta spectral amplitude
in primary (O1, O2) and associative (T5, T6) cortical
visual areas. At the same time, subjects suffering
from ASD, demonstrated significantly higher absolute
theta spectral amplitudes in the left frontal pole
region (Fp1) in the evening wakefulness, but not
in the morning wakefulness [48]. In another study,
Oberman et al. [49] recorded EEG signals from
10 high-functioning individuals of different ages
(6-47 years) and gender with ASD and from matched
control subjects in a parallel manner with watching
videos of a moving hand or a bouncing ball, with
moving their own hand, or against the background
of visual noise. Then the mu-frequency power at
scalp locations corresponding to sensorimotor cortex
(C3, Cz, and C4) during the self-initiated action and
watching action conditions was compared to the
power under baseline (visual white noise) conditions.
In other words, the mu wave suppression was
estimated. Ultimately, the report revealed that mu
suppression of self and observed hand movements
in control participants was significant. Meanwhile,
mu suppression of the participants with ASD in self-
induced hand movements (but not in observed hand
movements) was significant [49]. Stroganova et al. [50]
recorded EEG signals from 44 boys with ASD (ages
3-8 years) and from a corresponding number of age-
matched typically developing boys under conditions
including sustained visual attention (presentation of
soap bubbles or computer presentation of a moving
fish). Then, the authors evaluated the EEG spectral
power (SP) and SP interhemispheric asymmetry
within delta, theta, and alpha bands in both groups.
The researchers concluded that boys suffering from
ASD were in fact a heterogeneous group with regard
to the alpha and theta SPs. The group yielded a high
intergroup difference in absolute SPs of the prefrontal
delta. The left-side broadband EEG asymmetry in such
children was not typical, and the mid-temporal regions
had the maximum intensity in this regard [50].
Murias et al. [51] investigated EEG measures
in 18 male adults with ASD and 18 control male
subiects (18-38 years old) in the resting state with
the eyes closed. The coherence between pairs of
electrodes and the relative SPs were evaluated. For
the ASD group, locally higher coherence was clear
within the theta frequency range (3-6 Hz). This was
specifically evident in the temporal and frontal regions
of the left hemisphere. For the lower alpha range
(8-10 Hz), there was generally reduced coherence for
the ASD group in the frontal regions and also between
the frontal and all other scalp regions. In the ASD
group, the relative SPs of the ranges between 3-6 and
13-17 Hz were significantly higher, but this parameter
was significantly lower for 9-10 Hz [51].
Moreover, Bernier et al. [52] examined EEGs of
18 high-functioning adults with ASD and 15 IQ-
and age-matched typical subiects participated in
four conditions (resting, observation, execution, and
imitate). The EEG mu rhythm was compared between
two groups. The experiments illustrated that, with
executing an action, both groups exhibited significant
attenuation of the mu rhythm. When observing a
movement, however, considerably reduced attenuation
of the mu waves was observed in the ASD subjects
[52]. In another study, Orekhova et al. [53] recorded
EEGs in two independent samples of 3- to 8-year old
boys with autism (BWA) from Moscow (20 boys)
and Gothenburg (20 boys), and in the same number
of age-matched typically developing boys (TDB)
during sustained visual attention. The mean ASPs
were calculated for three high-frequency bands, beta
(13.2-24 Hz), gamma 1 (24.4-44.0 Hz), and gamma 2
(56.0-70 Hz). The authors reported that, a pathological
rise was observed in the gamma intensity (24.4-44.0
Hz) in both BWA samples. Also, there was positive
correlation between the intensity of gamma activity
and the developmental delay rate in both BWA
subgroups [53]. Coben et al. [54] investigated the
EEG measures under eyes-closed resting conditions in
20 children diagnosed with ASD and 20 controls
matched for gender, age (6-11 years old), and IQ. The
ASPs, RSPs, and total SPs, as well as intrahemispheric
and interhemispheric coherences, were calculated for
these two groups. It was found that children with ASD
noticeably differed in terms of the power and inter-
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2202
M. HASHEMIAN and H. POURGHASSEM
intrahemispheric coherences. In autistic children, an
excess theta-RSP appeared especially in the posterior
regions of the right hemisphere. Meanwhile, the
delta SP in the frontal cortex was rather low, but the
midline beta power was high. In the frontal regions
of both hemispheres, theta and delta coherences were
rather small. In addition, theta, delta, and alpha hypo-
coherences were observable in the temporal regions.
Finally, theta, delta, and beta coherences were rather
weak in the posterior regions [54].
Martineau et al. [55] compared EEG activity in
14 right-handed children with ASD and 14 right
handed, age- and gender-matched control children
(3 girls and 11 boys aged 5 years 3 months to 7 years
11 months) in the movie watching state. The silent
movie consisted of four sequences, namely (i) no
stimulation, “white” (Wh, TV screen white), (ii) a
no movement sequence, “lake” (Lk, a body of water
surrounded by land), (iii) a non-human movement
sequence, “waterfall” (Wf, falling water), and (iv) a
human movement sequence, “rotating” (Ro, a woman
performing scissor movements with her legs while
lying on her back). In this research, the logarithm
absolute spectral power (Ln ASP) in each of the
three following frequency bands, theta 1 (3-5.5 Hz),
theta 2 (5.5-7.5Hz), and alpha 1 (7.5-10.5 Hz), were
calculated. In was reported that, during observation
of human actions in the normal children group, EEG
desynchronization was observable in the frontal
and temporal cortices and in the motor cortex areas.
However, such a desynchronization was not evident in
ASD children [55].
Raymaekers et al. [36] investigated the mirror
neuron functioning. The EEG signals were recorded
from 20 children with high-functioning autism (HFA,
ages 8-13 years) and a control group of 19 typically
developing age-matched children. The testing
was based on the paradigm of Oberman et al. [49]
consisted of four conditions, (i) observing a video of a
moving hand (hand), (ii) moving own hand (self), (iii)
watching a video of two bouncing balls (balls), and
(iv) watching visual white noise (baseline). The mu
wave suppression was calculated as a ratio of 8-13 Hz
SP during each of the self, hand, and balls conditions
relative to the respective power under baseline
conditions. The report indicated that significant mu
suppression in both self and observed hand movements
were evident in both groups [36]. In the same year,
Lazarev et al. [56] investigated the EEG photic driving
at various stimulation frequencies (intermittent photic
stimulation at 11 fixed frequencies, from 3 to 24 sec–1
in 14 autistic boys (6-14 years old) and 21 control
boys matched in age. The interhemispheric asymmetry
in the total number of driving peaks in each group and
the difference between autistic and control groups in
each hemisphere were evaluated for each frequency
band of the four harmonics in the non-visual areas
and the sum of four harmonics in both non-visual and
occipital visual areas. The researchers deduced that
boys with autism showed latency abnormalities in the
right hemisphere during the photic driving reactivity,
particularly at the rapid alpha and beta frequencies of
stimulation [56].
Thatcher et al. [57] recorded EEG from 54 autistic
subjects and 241 normal subjects (2.6 to 11 years old)
under resting conditions with the eyes open. The EEG
phase shift and phase lock durations were computed
for all possible electrode combinations; two alpha
(8-10 and 10-13 Hz and three beta subranges 13-15, 15-
18, and 25-30 Hz) were considered. It was recognized
that the phase shift duration in ASD children in
both short (6 cm) and long (21-24 cm) interelectrode
distances in all bands of frequency, particularly in the
alpha1 frequency band (8-10 Hz), was significantly
shorter [57]. Chan et al. [58] investigated EEGs of
38 normal children and 16 children with ASD
(6-14 years old) that were recorded under the eyes-
open condition. Cordance was computed for 19
electrode sites using a three-step algorithm [44, 46]
as a feature. The obtained results also demonstrated
that cordance patterns of the ASD subjects were lower
as compared with those of normal ones, possibly
indicating that perfusion within the frontal regions of
ASD subjects is lower than that in normal ones [58].
Lazarev et al. [59] examined photic driving
coherence during intermittent photic stimulation in
14 autistic boys (6-14 years old, with IQ 91.4 ± 22.8)
and 19 normally developing boys who were subjected
to stimulation of 12 fixed frequencies (3 to 27 sec–1).
The number of high-coherent connections (HCC)
(coherence > 0.6-0.8) was estimated among seven
leads in each hemisphere. The findings showed
that, unlike the spectral characteristics indicating
deficit in the photic driving reactivity in the right
hemisphere, the groups were different in terms of the
number of HCC only in the left hemisphere. Also,
there was increased prevalence of the frequency in
the left hemisphere [59]. In the same year, Sudirman
et al. [60] collected EEGs from 6 normal children,
2 autism-suffering children, and 8 Down syndrome
children under the actions of two stimulus consisting
of alternating checkerboard and ripple checkerboard.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 203
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
203
The amplitudes of alpha frequency oscillations were
compared in the above three groups. It was found that
the alpha value for normal children, in comparison
with Down syndrome and autistic children, was higher
at 10 Hz [60].
Isler et al. [61] compared EEG activity in 6 children
with ASD and 8 age- and gender-matched control
children (5.5-8.5 years old) under visual stimulation
(long-latency flash-evoked visual potentials). The
EEG power and synchrony measures (coherence and
phase synchrony) were computed. In autistic children,
as compared with normal ones, the interhemisphere
synchrony demonstrated a 50% reduction in the theta
band. Also, the synchrony between the hemispheres
in autistic subjects was not distinguishable above the
theta band (uncorrelated cortical activity). In spite of
a power bilateral increase, the synchrony between the
hemispheres mitigated in autistic children. The wavelet
power in children with autism had a more sluggish
recovery, a faster primary reaction to stimulation, and
a more modularity state at longer latencies. Catarino
et al. [47], however, assessed EEGs in 15 participants
with ASD (23.79-42.34 years old) and 15 typical
controls (21.50-37.77 years old) under a face and
chair matching task (stimuli consisted of 30 pictures
of neutral faces and 30 pictures of chairs.) The multi-
scale entropy and RSPs were compared in two groups.
It was found that, in ASD children as compared to the
control group, the EEG signal complexity was lower
in the occipital and temporo-parietal areas. There
was no significant variation between the groups in
terms of the EEG power spectra [47]. In the same
year, Chan et al. [62] studied EEGs of 21 children
with ASD and 21 children with normal development
(5-14 years old) facing the object recognition task
(consisted of 24 line drawings taken from the object
database by Snodgrass et al. [63] and modified/
validated by Rossion et al. [64]). The line drawings
were placed in an array of six by four layouts
displayed on a computer screen for 3 min. The
participants were required to memorize the items
for a later recognition task (consisting of 12 targets
mixed with 12 distracters). In this research, theta
coherence measures (4-7.5 Hz) were used to evaluate
and analyze EEG signals. The authors [62] deduced
that ASD children, in comparison with normal ones,
demonstrated a dissimilar pattern of EEG coherence.
In 2012, two studies related to ASG/EEG were
published. Firstly, Lushchekina et al. [65] studied EEGs
in 5- to 7-year-old children, both normal and with early
childhood autism, under two resting conditions and at
a cognitive task. The SPs and mean coherence for the
alpha, beta, and gamma rhythms were compared. It
was mentioned that, for both ASD and normal children,
a frontal-occipital alpha gradient was considerable.
For normal children, the SP and coherence of EEG
rapid rhythms were significantly enhanced in the
frontal and central regions of the left hemisphere
under conditions of the cognitive tasks compared
with the baseline ones [65]. Secondly, Mathewson
et al. [66] investigated EEGs in 15 adults with ASD
(18.8-51.6 years old) and a matched comparison group
of 16 unimpaired adults (22.6-47.8 years old) under
eyes-closed and eyes-open conditions. The EEG alpha
SPs and coherence were computed for assessing the
participants. Calculations showed that there was a
difference between two groups in terms of coherence
or eyes-closed EEG alpha SP. However, alpha
suppression for eyes-open conditions was weaker in
ASD adults, as compared with normal ones.
ASD ANALYSIS BASED ON PATTERN
RECOGNITION TECHNIQUES
It has taken a giant leap in the direction of diagnosing
ASD based on EEG analysis. The researches have
used pattern recognition techniques to separate
ASD and non-ASD brain signal patterns. Figure 4
illustrates a general structure of the ASD diagnosis
algorithms based on the above-mentioned techniques.
These algorithms consist of two main components,
feature extraction and feature classification. The most
commonly utilized tools and approaches in these
algorithms are described below in detail.
Feature Extraction. The feature extraction stage
can be considered as a mapping from the initial signal
space to the feature space in a way that the separability
may be improved in the new space. Different features
EEG signals
(ASD and non-ASD)
ASD
Non-ASD
Preprocessing
(Filter, sampling)
Feature
extraction
Classification
F i g. 4. General structure of diagnostic algorithms based on pattern
recognition techniques.
Р и с. 4. Загальна структура діагностичних алгоритмів, що
базуються на техніці розпізнавання образів.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2204
M. HASHEMIAN and H. POURGHASSEM
are extracted by certain methods and scenarios from
the EEG signals. Eventually, the extracted features
form a vector that is called the “feature vector.” The
feature vectors of samples are used in the classification
stage. Table 1 exhibits the effective features that were
extracted from the frequency and time domains of
EEG signals in the ASD diagnosis algorithms.
Classification. Classification is the process of
assigning a feature vector to one of predefined classes
or categories in a manner that minimizes the error of
classification [67]. In the ASD detection problem, two
classes are individuals with ASD (class one) and those
with no ASD (class two). This process is usually done
by applying classifiers on the feature vectors of two
classes. The essential part of this process is to know
how a classifier assigns one of the two classes to an
unknown feature vector. Almost all of the classifiers
have a training phase by a specific algorithm, and
then they become capable of classifying the samples.
In actuality, the above-mentioned specific algorithms
use feature vectors that have previously been extracted
from EEG signals of both groups. There are varieties
of classifiers, like an artificial neural network,
support vector machine (SVM), statistical classifier,
K-Nearest Neighbor (KNN), Linear Discriminant
Analysis (LDA), and Quadratic Discriminant Analysis
(QDA). In fact, each one of them has its own strategy.
Parameter setting to each classifier would have a
direct impact on its performance. We could easily see
the significance of feature extraction process in the
diagnostic algorithms because the input of classifiers
is formed by feature vectors.
Algorithms Based on Pattern Recognition
Techniques. In this section, we would like to describe
the algorithms based on pattern recognition techniques
provided by the researchers in the ASD detection.
Here, we have tried to bring up important factors in
each algorithm, including the extracted features, the
methods employed, the type of utilized classifier,
and that of database. The database implies the EEG
signals of ASD and non-ASD individuals, which have
been used in the process of designing and evaluating
the algorithm. Each database has its own parameters,
such as conditions of EEG recording, the number
of ASD and non-ASD people, their age range, and
their IQ. For example, Sheikhani et al. [68] used
EEG samples of 11 patients (9.2 ± 1.4 years old) and
10 control age-matched subjects under eye-opened
conditions. The Lempel-Ziv (LZ) complexity, Short
Time Fourier Transform (STFT), and STFT at a
bandwidth (STFT-BW) in the total spectrum were
extracted from EEG signals and then evaluated by the
ANOVA test. Finally, the STFT-BW feature obtained
the most difference between these two groups on the
basis of ANOVA. In this study, the KNN classifier has
been used to classify a feature vector. This algorithm
has obtained 81.0% discrimination between normal
and autism subjects with Mahalanobis distance [68]. In
another study, Behnam et al. [69] utilized EEG signals
of 10 ASD (6-11 years old) and 9 age-matched control
subjects, which were collected under eye-opened
conditions. The STFT-BW component in the alpha band
(8-12 Hz) was calculated as a feature, and a KNN
classifier with Mahalanobis distance was used.
Eventually, this algorithm was able to separate the
normal peoples from ASD ones with the accuracy rate
of 89.5%. In this study, moreover providing diagnostic
algorithm, the coherence measures between all
171 pairs of 19 channels in three frequency bands
(alpha, beta, and gamma) of EEG were examined. The
authors declared that there are more abnormalities in
the connectivity between the left hemisphere and right
The Employed Features of EEG Signals in the Frequency and Time Domains
Частотні та часові характеристики ЕЕГ-сигналів, застосовані для аналізу
Frequency/time domains Description
Frequency domain
Short-Time Fourier Transform at a bandwidth (STFT-BW) in the total spectrum [68]
STFT-BW component in the alpha band [69]
Averaged values of the spectrogram greater than 70% maximum in the alpha frequency band [70],[35]
Principle Components Analysis (PCA) to Short-Time Fourier Transform [74]
Gaussian mixture model (GMM) in frequency domain [72]
Katz’s Fractal Dimensions in delta and gamma EEG sub-bands [71]
Principal Components Analysis (PCA) of the coherence data [75]
The raw data and Fast Fourier Transform ( FFT) [76]
Time domain Modified multi-scale entropy (mMSE) [73]
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 205
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
temporal lobe, as compared with other regions [69].
Sheikhani et al. [70] utilized qEEG of 15 children
with Asperger disorder (10 boys and 5 girls, age
6-11 years) and 11 normal children (7 boys and
4 girls of the same age range). The EEG signals of
two groups of subjects were recorded under nine
conditions, including the eye-closed state, relaxed
eye-opened condition, looking at three samples of the
Kanizsa puzzle, looking at a mother’s picture upright
and inverted, and looking at a stranger’s picture
upright and inverted. The average spectrogram values
greater than 70% of the maximum were employed
as a discriminating feature on quantitative EEG
signals in the frequency bands of delta (0-4 Hz), theta
(4-8 Hz), alpha (8-12 Hz), beta (12-36 Hz), and gamma
(36-44 Hz). For classification of ASD children vs
normal children, a KNN classifier with Mahalanobis
distance was utilized. Experimental results showed
that the recorded signals under relaxed open-eyed
conditions in the gamma band, those recorded with
looking at a stranger’s inverted-condition picture in
the alpha and beta bands, and the ones obtained with
participants looking at a mother’s inverted picture in
the beta band provided the best discriminations with
the accuracy rate of 96.2, 83.3, 70.6, and 77.8%,
respectively [70]. In the next year, Sheikhani et
al. [35] gathered qEEG signals from 17 children
(13 boys and 4 girls, 6 to 11 years) with ASD and 11
control children (7 boys and 4 girls of the same age
range) under relaxed eye-opened conditions. Average
values of the spectrogram (STFT) greater than 70%
maximum (spectrogram criteria) were calculated from
quantitative EEG signals in the delta, theta, alpha,
beta, and gamma frequency bands. Among the obtained
amounts in each frequency band, average values of the
spectrogram in the alpha band showed the maximum
difference between two groups, and such value was
chosen as a feature. Finally, this algorithm was able
to differentiate sick and healthy (control group)
individuals with the accuracy rate of 96.4% [35].
Ahmadlou et al. [71] collected EEG signals from
9 ASD children (6 to 13 years old, average age of
10.8 years), and 8 non-ASD children (7 to 13 years old,
average age of 11.2 years) under resting eye-closed
conditions. Then, the Higuchi’s fractal dimension
(FD) and Katz’s fractal dimension were computed
in all EEG subranges produced by the wavelet
decomposition, as well as in the entire band-limited
EEG. Significant FDs in different loci and different
EEG subranges or band-limited EEG for distinguishing
ASD children from non-ASD children were determined
by ANOVA. Finally, three characteristics, including
Katz’s fractal dimensions in delta (of loci Fp2 and C3)
and gamma (of locus T6) EEG sub-bands, were chosen
among the extracted features by ANOVA. The EEG
data are classified into ASD and non-ASD children
groups using the radial basis function classifier
(RBFNN). This classifier yielded the accuracy rate
of 90.0% for diagnosis of the ASD in the three-
dimensional feature space [71].
In 2011, two main studies were reported. Firstly,
Razali et al. [72] used EEG signals from 6 autistic and
6 control children (each group with the age around 7 to
9 years old) under conditions of a motor imitation task
(to clinch their hand according to the video stimuli).
A Gaussian mixture model was used as a method of
feature extraction for analyzing the brain signals in
the frequency domain. Then, the extraction data were
classified using Multilayer Perceptron (MLP). This
algorithm acquired 86.62% discrimination between
two groups [72]. Secondly, Bosl et al. [73] collected
an EEG database from 79 different infants (46 high
ASD-risk infants, HRA, and 33 controls of five age
groups, 6, 9, 12, 18, and 24 months) under resting state
conditions. Modified Multi-scale Entropy (mMSE)
was extracted as a feature vector. To obtain the best
classification, the authors examined operations of
three types of classifiers, including KNN, Bayes, and
SVM. The differences appeared to be the greatest at
ages of 9 to 12 months. In a nutshell, infants were
classified with an over-80% accuracy into control and
HRA groups at the age of 9 months. The classification
accuracy for boys was close to 100% and remained
high (70 to 90%) at the ages of 12 and 18 months. For
girls, the classification accuracy was highest at the age
of 6 months but declined thereafter [73].
Khazaal Shams et al . [74] collected EEG
signals from six autistic children and six typical
preschool subjects (around 7 to 9 years old) under
two conditions in the open-eyed state and motor
task movement (asked to follow the right and left
hand movement movie). The feature extraction was
performed by Principle Components Analysis (PCA)
to STFT of the EEG signals. Then, MLP is used to
classify the feature vectors. The results showed that
the proposed algorithm gives the accuracy rate of
90-100% for autism and normal children in the motor
task and around 90% in the detection of normal
subjects in the open-eyed task [74].
In another study, Duffy et al. [75] gathered EEG data
from 463 children who were diagnosed with ASD and
from 571children considered neurotypical controls,
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2206
M. HASHEMIAN and H. POURGHASSEM
with ages ranging from 1 to 18 years, in the awake
and alert state. The spectral coherence was calculated;
then PCA of the coherence data was employed as an
objective technique to reduce the variable number
meaningfully. For 2- to 12-year-old children,
40 factors of PCA showed highly significant intergroup
differences (P < 0.0001). Discriminant function
analysis (DFA) was used for classification that
yielded precision of 88.5% for the control subjects and
precision of 86.0% for the individuals with ASD [75].
Alhaddad et al. [76] gathered EEG samples from eight
children with ASD (5 boys and 3 girls, 10-11 years
old) and four control children (all of them were boys,
10-11 years old) under a relaxed condition. The authors
investigated different preprocessing techniques, such
as referencing, filtering, windsorizing, and scaling,
for obtaining the best classification accuracy. After
preprocessing, the raw data and FFT were used as
features. Finally, the extracted features were classified
using Fisher Linear Discriminant Analysis (FLDA). It
was reported that, among the applied preprocessing
techniques, the Windsor-filtered data gave the best
performance for both raw data (89.97 ± .02%) and FFT
features (91.64 ± .021%) [76].
Evaluation Measures of the Diagnosis Algorithm.
As was previously mentioned, the designed classifier
in the diagnosis algorithms is trained by a dataset,
and then the trained classifier is able to assign any
unknown sample to either ASD class or non-ASD
class. In the reality, after the design of an algorithm, it
will be examined by a test dataset having samples of
the ASD and non-ASD individuals already diagnosed
by physicians. Now, the EEG signals of the individuals
enter the algorithm (after preprocessing and feature
extraction), and eventually the algorithm assigns one
of the two labels (ASD and non-ASD) to the EEG
signal. If the output of algorithm matched the findings
of physicians, the sample has been classified correctly.
In other words, the algorithm performance is calculated
as the number of test samples identified correctly by
the algorithm to the total number of the test dataset.
Based on the above finding, an important question
comes to mind: On what size of the test dataset the
performance of the algorithm is based? In fact, the size
of the test dataset should be large enough until the
used classifier can generalize the unknown samples. It
seems that the bigger size of the test dataset causes the
higher validation of the performance. In other words,
the algorithm enjoys higher generalization. However,
due to the limitations of collecting the test dataset, the
latter is usually rather small in most researches. Now,
another question may be raised: How we can achieve
a high generalization algorithm in a limited test
dataset? The cross-validation methods are the answer
to this question. By using these methods in designing
diagnostic algorithm, a highly reliable performance
could be obtained despite of small database. Some
of these methods are random subsampling, k-fold,
and leave one-out. Some researchers have used these
methods in their proposed ASD diagnostic algorithms.
CONCLUSION
We proposed a surway of the studies on ASD diagnosis
algorithms based on EEG analysis. We found that the
studies could be divided into two groups, analysis
based on comparison techniques and analysis based
on pattern recognition techniques. Analysis based
on comparison techniques has been able to identify,
by using statistical methods, some of the features of
EEG, which were different in the ASD and non-ASD
individuals. Through reviewing the data of the studies,
we found that the results of these studies are dissimilar.
Analysis based on pattern recognition techniques
takes a big step in diagnosing ASD based on EEG
signals. In such studies, the researchers were able to
take the advantage of pattern recognition techniques
to differentiate the brain signal patterns affected
by ASD from those of non-ASD ones. The feature
extraction and classification are two main components
in the structure of all these algorithms. In the feature
extraction phase, various features with different
scenarios are extracted from EEG signals. Among all
extracted features, the ones that highlight the greatest
difference between two groups are selected and used
in designing the algorithm. Eventually, the classifiers
assign a label of either ASD or non-ASD to the EEG
signals by using the extracted features. Each one of
the diagnostic algorithms reports a performance based
on a test dataset. It should be noted that the degree of
generalization and validity of a diagnostic algorithm
depend on two factors, the size of the used test dataset
and the type of utilized cross-validation methods.
М. Хашемян1, Х. Пургассем1
ДІАГНОСТИКА РОЗЛАДІВ АУТИСТИЧНОГО СПЕКТРА,
ЩО ҐРУНТУЄТЬСЯ НА АНАЛІЗІ ЕЕГ (огляд)
1 Наджафабадський підрозділ Ісламського університету
Азад, Ісфаган (Іран).
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 207
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
Р е з ю м е
Розлади аутистичного спектра (autism spectrum disorders –
ASD) – це глибокі відхилення розвитку нервової
сфери, що характеризуються порушенням соціальних
взаємодій, комунікативних навичок та стереотипної
поведінки. Оскільки реєстрація та аналіз ЕЕГ є одними
із фундаментальних засобів діагностики та ідентифікації
нейрофізіологічних розладів, дослідники намагаються
використовувати ЕЕГ-сигнали для діагностики ASD у тих
або інших осіб. Як ми встановили, дослідження, спрямовані
на діагностику ASD із застосуванням ЕЕГ-методик, можуть
бути поділені на дві групи, коли аналіз базується або на
техніці порівнянь, або на техніці розпізнавання образів.
У цьому огляді ми намагались описати застосування двох
відповідних комплексів алгоритмів, а також методики
їх використання та отримані результати. Нарешті,
обговорюється порівняльна ефективність вказаних
алгоритмів діагностування.
REFERENCES
1. J. L. Matson and J. A. Boisijoli, “Strategies for assessing
Asperger’s syndrome: A critical review of data based methods,”
Res. Autism Spectr. Disord., 2, No. 2, 237-248 (2008).
2. J. L. Matson and N. F. Minshawi, Early Intervention for
Autism Spectrum Disorders: A Critical Analysis, Elsevier Sci.
Inc., Oxford (2006).
3. M. Hashemian and H. Pourghassem, “Facial emotion
processing in autism spectrum disorder based on spectral
features of EEG signals,” Int. J. Imaging Robotics, 11, No. 3,
68-80 (2013).
4. C. M. Zaroff and S. Y. Uhm, “Prevalence of autism spectrum
disorders and influence of country of measurement and
ethnicity,” Soc. Psychiat. Psychiatr. Epidemiol., 47, No. 3,
395-398 (2012).
5. S. Baron-Cohen, F. J. Scott, C. Allison, et al., “Prevalence
of autism-spectrum conditions: UK school-based population
study,” Br. J. Psychiat., 194, No. 6, 500-509 (2009).
6. T. S. Brugha, S. McManus, J. Bankart, et al., “Epidemiology
of autism spectrum disorders in adults in the community in
England,” Arch. Gen. Psychiat., 68, No. 5, 459-465 (2011).
7. B. Mulvihill, M. Wintage, R. S. Kiby, et al., “Prevalence
of autism spectrum disorders – Autism and Developmental
Disabilities Monitoring Network, United States, 2006,”
MMWR Surveill Summ., 58, No. 10, 1-20 (2009).
8. E. Fombonne, “Epidemiology of pervasive developmental
disorders,” Pediat. Res., 65, No. 6, 591-598 (2009).
9. A. Ghanizadeh, “A preliminary study on screening prevalence
of pervasive developmental disorder in school children in
Iran,” J. Autism. Dev. Disord., 38, No. 4, 759-763 (2008).
10. A. A. Nejatisafa, M.R. Kazemi, and J. Alaghebandrad,
“Autistic features in adult population: evidence for continuity
of autistic symptoms with normality,” Adv. Cognitive Sci., 5,
No. 3, 34-39 (2003).
11. M. Othman and A. Wahab, “Affective face processing analysis
in autism using electroencephalogram,” in: Proceedings of the
International Conference on Information and Communication
Technology for the Muslim World (ICT4M) (December 13-14,
2010), Jakarta (2010).
12. W. Klimesch, “EEG alpha and theta oscillations reflect
cognitive and memory performance: a review and analysis,”
Brain Res. Brain Res. Rev., 29, Nos. 2/3, 169-195 (1999).
13. S. Nasehi and H. Pourghassem, “Online mental task
classification based on DWT-PCA features and probabilistic
neural network,” Int. J. Imaging Robotics, 7, No. 1, 110-118
(2012).
14. S. Nasehi and H. Pourghassem, “A novel fast epileptic seizure
onset detection algorithm using general tensor discriminant
analysis,” J. Clin. Neurophysiol., 30, No. 4, 362-370 (2013).
15. W. Klimesch, G. Pfurtscheller, and H. Schimke, “Pre- and post-
stimulus processes in category judgement tasks as measured by
event-related desynchronization (ERD),” J. Psychophysiol., 6,
185-203 (1992).
16. W. Klimesch, “Memory processes, brain oscillations and EEG
synchronization,” Int. J. Psychophysiol., 24, Nos. 1/2, 61-100
(1996).
17. W. Klimesch, H. Schimke, and J. Schwaiger, “Episodic and
semantic memory: An analysis in the EEG theta and alpha
band,” Electroencephalogr. Clin. Neurophysiol., 91, No. 6,
428-441 (1994).
18. S. Nasehi and H. Pourghassem, “Automatic prediction of
epileptic seizure using kernel fisher discriminant classifiers,”
in: Proceeding of the International Conference on Intelligent
Computation and Bio-Medical Instrumentation (ICBMI),
(14-17 Dec. 2011), Wuhan, China (2011), pp. 200-203.
19. W. Klimesch, M. Doppelmayr, D. Rohm, et al., “Simultaneous
desynchronization and synchronization of different alpha
responses in the human electroencephalograph: a neglected
paradox,” Neurosci. Lett., 284, Nos. 1/2, 97-100 (2000).
20. J. A. Pineda, “The functional significance of mu rhythms:
Translating ‘seeing’ and ‘hearing’ into ‘doing’,” Brain Res.
Rev., 50, No. 1, 57-68 (2005).
21. S. Salenius, A. Schnitzler, R. Salmelin, et al., “Modulation of
human cortical rolandic rhythms during natural sensorimotor
tasks,” NeuroImage, 5, No. 3, 221-228 (1997).
22. R . Ha r i , R . Sa lme l in , J . P. Mäke l ä , e t a l . ,
“Magnetoencepha lographic cor t ica l rhy thms,” In t .
J. Psychophysiol., 26, Nos. 1/3, 51-62 (1997).
23. G. Pfurtscheller, “Central beta rhythm during sensorimotor
activities in man,” Electroencephalogr. Clin. Neurophysiol.,
51, No. 3, 253-264 (1981).
24. G. Pfurtscheller, A. Stancak, and C. Neuper, “Post-movement
beta synchronization. A correlate of an idling motor area,”
Electroencephalogr. Clin. Neurophysiol., 98, No. 4, 281-293
(1996).
25. G. Pfurtscheller, K. Zalaudek, and C. Neuper, “Event-related
beta synchronization after wrist, finger and thumb move-
ment,” Electroencephalogr. Clin. Neurophysiol., 109, No. 2,
154-160 (1998).
26. E. Basar, C. Basar-Eroglu, S. Karakas, et al., “Brain
oscillations in perception and memory,” Int. J. Psychophysiol.,
35, Nos. 2/3, 95-124 (2000).
27. C. Tallon-Baudry, O. Bertrand, C. Delpuech, et al., “Stimulus
specificity of phase-locked and non-phase-locked 40 Hz visual
responses in human,” J. Neurosci., 16, No. 13, 4240-4249
(1996).
28. C. Tallon-Baudry, O. Bertrand, C. Wienbruch, et al.,
“Combined EEG and MEG recordings of visual 40 Hz
responses to illusory triangles in human,” NeuroReport, 8,
No. 5, 1103-1107 (1997).
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2208
M. HASHEMIAN and H. POURGHASSEM
29. C. Tallon-Baudry, O. Bertrand, F. Peronnet, et al., “Induced
gamma-band activity during the delay of a visual short- term
memory task in humans,” J. Neurosci., 18, No. 11, 4244-4254
(1998).
30. C. M. Krause, P. Korpilahti, B. Porn, et al., “Automatic
auditory word perception as measured by 40 Hz EEG
responses,” Electroencephalogr. Clin. Neurophysiol., 107,
No. 2, 84-87 (1998).
31. N. E. Crone, D. L. Miglioretti , B. Gordon, et al . ,
“Functional mapping of human sensorimotor cortex with
electrocorticographic spectral analysis. II. Event-related
synchronization in the gamma band,” Brain, 121, No. 12,
2301-2315 ( 1998)
32. P. L. Nunez, R. Srinivasan, A. F. Westdorp, et al., “EEG
coherency. I . Statistics, reference electrode, volume
conduction, Laplacians, cortical imaging, and interpretation
at multiple scales,” Electroencephalogr. Clini. Neurophysiol.,
103, No. 5, 499-515 (1997).
33. G. Nolte, O. Bai, L. Wheaton, et al., “Identifying true brain
interaction from EEG data using the imaginary part of
coherency,” Clin. Neurophysiol., 115, 2292-2307 (2004).
34. E. Pereda, R. Q. Quiroga, and J. Bhattacharya, “Nonlinear
multivariate analysis of neurophysiological signals,” Prog.
Neurobiol., 77, Nos. 1/2, 1-37 (2005).
35. A. Sheikhani, H. Behnam, M. R. Mohammadi, et al.,
“Detection of abnormalities for diagnosing of children with
autism disorders using of quantitative electroencephalography
analysis,” J. Med. Syst., 36, No. 2, 957-963 (2010).
36. R. Raymaekers, J. R. Wiersema, and H. Roeyers, “EEG study
of the mirror neuron system in children with high functioning
autism,” Brain Res., 1304, 113-121 (2009).
37. C. Babiloni, F. Babiloni, F. Carducci, et al., “Human cortical
electroencephalography (EEG) rhythms during the observation
of simple aimless movements: a high-resolution EEG study,”
NeuroImage, 17, No. 2, 559-572 (2002).
38. J. A. Pineda, B. Z. Allison, A. Vankov, “The effects of self-
movement, observation and imagination on mu rhythms and
readiness potentials (RP’s): toward a brain-computer interface
(BCI),” IEEE Trans. Rehabil. Eng., 8, No. 2, 219-222 (2000).
39. S. Cochin, C. Barthelemy, S. Roux, et al., “Observation
and execution of movement: Similarities demonstrated by
quantified electroencephalography,” Eur. J. Neurosci., 11,
No. 5, 1839-1842 (1999).
40. A. F. Leuchter, S. H. J. Uijtdehaage, I. A. Cook, et al.,
“Relationship between brain electrical activity and cortical
perfusion in normal subjects,” Psychiat. Res. (Neuroimaging
Sect.), 90, 125-140 (1999).
41. I. A. Cook, R. O’Hara, S. H. J. Uijtdehaage, et al., “Assessing
the accuracy of topographic EEG mapping for determining
local brain function,” Electroencephalogr. Clin. Neurophysiol.,
107, No. 6, 408-414 (1998b).
42. I. A. Cook, A. F. Leuchter, M. L. Morgan, et al., “Early
changes in prefrontal activity characterize clinical responders
to antidepressants,” Neuropsychopharmacology, 27, 120-131
(2002).
43. M. Costa, A. L. Goldberger, and C. K. Peng, “Multiscale
entropy analysis of complex physiologic time series,” Phys.
Rev. Lett., 89, No. 6, 068102 (2002).
44. M. Costa, A. L. Goldberger, and C. K. Peng, “Multiscale
entropy analysis of biological signals,” Phys. Rev. E. Stat.
Nonlin. Soft. Matter Phys., 71, No. 2, 021906 (2005).
45. J. S. Richman, D. E. Lake, and J. R. Moorman, “Sample
entropy analysis,” Methods Enzymol., 384, 172-184 (2004).
46. J. S. Richman and J. R. Moorman, “Physiological time-series
analysis using approximate entropy and sample entropy,” Am.
J. Physiol. Heart Circ. Physiol., 278, No. 6, 2039-2049 (2000).
47. A. Catarino, O. Churches, S. Baron-Cohen, et al. “Atypical
EEG complexity in autism spectrum conditions: A multiscale
entropy analysis,” Clin. Neurophysiol., 122, 2375-2383 (2011).
48. A. M. Daoust, E. Limoges, C. Bolduc, et al., “EEG spectral
analysis of wakefulness and REM sleep in high functioning
autistic spectrum disorders,” Clin. Neurophysiol., 115, 1368-
1373 (2004).
49. L. M. Obermana, E. M. Hubband, J. P. Mccleery, et al., “EEG
evidence for mirror neuron dysfunction in autism spectrum
disorders,” Brain Res. Cogn. Brain Res., 24, No. 2, 190-198
(2005).
50. T. A. Stroganova, G. Nygren, M. M. Tsetlin, et al., “Abnormal
EEG lateralization in boys with autism,” Clin. Neurophysiol.,
118, 1842-1854 (2007).
51. M. Murias, S. J. Webb, J. Greenson, et al., “Resting state
cortical connectivity reflected in EEG coherence in individuals
with autism,” Biol. Psychiat., 62, No. 3, 270-273 (2007).
52. R. Bernier, G. Dawson, S. Webb, et al., “EEG mu rhythm and
imitation impairments in individuals with autism spectrum
disorder,” Brain Cogn., 64, 228-237 (2007).
53. E. V. Orekhova, T. A. Stroganova, G. Nygren, et al., “Excess of
high frequency electroencephalogram oscillations in boys with
autism,” Biol. Psychiat., 62, 1022-1029 (2007).
54. R. Coben, A. R. Clarke, W. Hudspeth, et al., “EEG power and
coherence in autistic spectrum disorder,” Clin. Neurophysiol.,
119, 1002-1009 (2008).
55. J. Martineau, S. Cochin, R. Magne, et al., “Impaired cortical
activation in autistic children: Is the mirror neuron system
involved?” Int. J. Psychophysiol., 68, 35-40 (2008).
56. V. V. Lazarev, A. Pontes, and L. C. deAzevedo, “EEG photic
driving: Right-hemisphere reactivity deficit in childhood
autism. A pilot study,” Int. J. Psychophysiol., 71, 177-183
(2009).
57. R. W. Thatcher, D. M. North, J. Neubrander, et al., “Autism
and EEG phase reset: deficient GABA mediated inhibition in
thalamo-cortical circuits,” Dev. Neuropsychol., 34, No. 6, 780-
800 (2009).
58. A. S. Chan, M. C. Cheung, Y. M. Han, et al., “Executive
function deficits and neural discordance in children with
autism spectrum disorders,” Clin. Neurophysiol., 120, No. 6,
1107-1115 (2009).
59. V. V. Lazarev, A. Pontes, A. A. Mitrofanov, et al. ,
“Interhemispheric asymmetry in EEG photic driving coherence
in childhood autism,” Clin. Neurophysiol., 121, 145-152
(2010).
60. S. Sudirman, S. Saidin, and N. M. Safri, “Study of
electroencephalography signal of autism and down syndrome
children using FFT,” in: Proceedings of the IEEE Symposium
on Industrial Electronics and Applications, Penang (2010),
pp. 401-406.
61. J. R. Isler, K. M. Martien, P. G. Grieve, et al., “Reduced
functional connectivity in visual evoked potentials in children
with autism spectrum disorder,” Clin. Neurophysiol., 121,
2035-2043 (2010).
62. A. S. Chan, Y. M. Y. Han, S. L. Sze, et al., “Disordered
connectivity associated with memory deficits in children with
autism spectrum disorders,” Res. Autism Spectr. Disord., 5,
237-245 (2011).
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 2 209
AUTISM SPECTRUM DISORDERS DIAGNOSING BASED ON EEG ANALYSIS
63. J. G. Snodgrass and M. Vanderwart, “A standardized set of
260 pictures: Norm for name agreement, image agreement,
familiarity, and visual complexity,” J. Exp. Psychol. Hum.
Learn., 6, 174-215 (1980).
64. B. Rossion and G. Pourtois, “Revisiting Snodgrass and
Vanderwart’s object pictorial set: The role of surface detail in
basic-level object recognition,” Perception, 33, No. 2, 217-236
(2004).
65. E. A. Lushchekina, E. D. Podreznaya, V. S. Lushchekin, et al.,
“A comparative EEG study in normal and autistic children,”
Neurosci. Behav. Physiol., 42, No. 3, 236-243 (2012).
66. K. J. Mathewson, M. K. Jetha, I. E. Drmic, et al., “Regional
EEG alpha power, coherence, and behavioral symptomatology
in autism spectrum disorder,” Clin. Neurophysiol., 123, No. 9,
1798-1809 (2012).
67. S. Nasehi and H. Pourghassem, “Seizure detection algorithms
based on analysis of EEG and ECG signals: A survey,”
Neurophysiology, 44, 174-186 (2012).
68. A. Sheikhani, H. Behnam, M. R. Mohammadi, et al., “Analysis
of quantitative electroencephalogram background activity in
autism disease patients with Lempel-Ziv complexity and Short-
Time Fourier Transform measure,” in: Proceedings of the 4th
IEEE/EMBS, International Summer School and Symposium
on Medical Devices and Biosensors. (August 19-22, 2007),
Cambridge (2007), pp. 19-22.
69. H. Behnam, A. Sheikhani , M. R. Mohammadi, e t
a l . , “Abnormali t ies in connect ivi ty of quanti tat ive
electroencephalogram background activity in autism disorders
especially in left hemisphere and right temporal,” in:
Proceedings of the 10th International Conference on Computer
Modeling and Simulation (April 1-3, 2008), Cambridge
(2008), pp. 82-87.
70. A. Sheikhani, H. Behnam, M. Noroozian, et al., “Abnormalities
of quantitative electroencephalography in children with
Asperger disorder in various conditions,” Res. Autism Spectr.
Disord., 3, 538-546 (2009).
71. M. Ahmadlou, H. Adeli, and A. Adeli, “Fractality and a
wavelet-chaos-neural network methodology for EEG-based
diagnosis of autistic spectrum disorder,” J. Clin. Neurophysiol.,
27, No. 5, 328-333 (2010).
72. N. Razali and A. Wahab, “2D affective space model (ASM) for
detecting autistic children,” in: Proceedings of the 15th IEEE
International Symposium on Consumer Electronics (June 14-
17, 2011), Singapore (2011), pp. 536-541.
73. W. Bosl, A. Tierney, H. Tager-Flusberg, et al., “EEG
complexity as a biomarker for autism spectrum disorder risk,”
BMC Medicine, 9, No. 18, (2011).
74. W. K. Shams and A. Wahab, “Characterizing autistic disorder
based on principle component analysis,” Austr. J. Basic Appl.
Sci., 6, No. 1, 149-155 (2012).
75. F. H. Duffy and H. Als, “A stable pattern of EEG spectral
coherence distinguishes children with autism from neurotypical
controls – a large case control study,” BMC Med., 10, No. 64
(2012); doi: 10. 1186/1784-7015-10.64.
76. M. J. Alhaddad, M. I. Kamel, H. M. Malibary, et al., “Diagnosis
autism by fisher linear discriminant analysis FLDA via EEG,”
Int. J. Bio-Sci. Bio-Technol., 4, No. 2 (2012).
|
| id | nasplib_isofts_kiev_ua-123456789-148271 |
| institution | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| issn | 0028-2561 |
| language | English |
| last_indexed | 2025-11-29T12:08:46Z |
| publishDate | 2014 |
| publisher | Інститут фізіології ім. О.О. Богомольця НАН України |
| record_format | dspace |
| spelling | Hashemian, M. Pourghassem, H. 2019-02-17T19:30:36Z 2019-02-17T19:30:36Z 2014 Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey / M. Hashemian, H. Pourghassem // Нейрофизиология. — 2014. — Т. 46, № 2. — С. 197-209. — Бібліогр.: 76 назв. — англ. 0028-2561 https://nasplib.isofts.kiev.ua/handle/123456789/148271 616.89+57.02+616-071 Autism spectrum disorders (ASD) are pervasive neurodevelopmental conditions characterized by impairments in reciprocal social interactions, communication skills, and stereotyped behavior. Since EEG recording and analysis is one of the fundamental tools in diagnosis and identifying disorders in neurophysiology, researchers strive to use the EEG signals for diagnosing of individuals with ASD. We found that studies on the ASD diagnosis using EEG techniques could be divided into two groups, where analysis was based on either comparison techniques or pattern recognition techniques. In this paper, we try to explain these two sets of algorithms along with their applied methods and results. Ultimately, evaluation measures of diagnosis algorithms are discussed Розлади аутистичного спектра (autism spectrum disorders – ASD) – це глибокі відхилення розвитку нервової сфери, що характеризуються порушенням соціальних взаємодій, комунікативних навичок та стереотипної поведінки. Оскільки реєстрація та аналіз ЕЕГ є одними із фундаментальних засобів діагностики та ідентифікації нейрофізіологічних розладів, дослідники намагаються використовувати ЕЕГ-сигнали для діагностики ASD у тих або інших осіб. Як ми встановили, дослідження, спрямовані на діагностику ASD із застосуванням ЕЕГ-методик, можуть бути поділені на дві групи, коли аналіз базується або на техніці порівнянь, або на техніці розпізнавання образів. У цьому огляді ми намагались описати застосування двох відповідних комплексів алгоритмів, а також методики їх використання та отримані результати. Нарешті, обговорюється порівняльна ефективність вказаних алгоритмів діагностування. en Інститут фізіології ім. О.О. Богомольця НАН України Нейрофизиология Обзоры Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey Діагностика розладів аутистичного спектра, що ґрунтується на аналізі ЕЕГ (огляд) Article published earlier |
| spellingShingle | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey Hashemian, M. Pourghassem, H. Обзоры |
| title | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey |
| title_alt | Діагностика розладів аутистичного спектра, що ґрунтується на аналізі ЕЕГ (огляд) |
| title_full | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey |
| title_fullStr | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey |
| title_full_unstemmed | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey |
| title_short | Autism Spectrum Disorders Diagnosing Based on EEG Analysis: a Survey |
| title_sort | autism spectrum disorders diagnosing based on eeg analysis: a survey |
| topic | Обзоры |
| topic_facet | Обзоры |
| url | https://nasplib.isofts.kiev.ua/handle/123456789/148271 |
| work_keys_str_mv | AT hashemianm autismspectrumdisordersdiagnosingbasedoneeganalysisasurvey AT pourghassemh autismspectrumdisordersdiagnosingbasedoneeganalysisasurvey AT hashemianm díagnostikarozladívautističnogospektraŝogruntuêtʹsânaanalízíeegoglâd AT pourghassemh díagnostikarozladívautističnogospektraŝogruntuêtʹsânaanalízíeegoglâd |