Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm
In this study, event-related potentials (ERPs) collected from normally hearing subjects and elicited by a multi-feature paradigm were investigated, and mismatch negativity (MMN) was detected. Standard stimuli and five types of deviant stimuli were presented in a specified sequence, while EEG dat...
Збережено в:
| Дата: | 2014 |
|---|---|
| Автори: | , , , , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут фізіології ім. О.О. Богомольця НАН України
2014
|
| Назва видання: | Нейрофизиология |
| Онлайн доступ: | http://dspace.nbuv.gov.ua/handle/123456789/148296 |
| 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: | Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm / M. Najafi-Koopaie, H. Sadjedi, S. Mahmoudian, E.D. Farahani, M. Mohebbi // Нейрофизиология. — 2014. — Т. 46, № 4. — С. 01-410. — Бібліогр.: 30 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-148296 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1482962019-02-18T01:25:12Z Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm Najafi-Koopaie, M. Sadjedi, H. Mahmoudian, S. Farahani, E.D. Mohebbi, M. In this study, event-related potentials (ERPs) collected from normally hearing subjects and elicited by a multi-feature paradigm were investigated, and mismatch negativity (MMN) was detected. Standard stimuli and five types of deviant stimuli were presented in a specified sequence, while EEG data were recorded digitally at a 1024 sec–1 sampling rate. Two wavelet analyses were compared with a traditional difference-wave (DW) method. The Reverse biorthogonal wavelet with an order of 6.8 and the quadratic B-Spline wavelet were applied for seven-level decomposition. The sixth-level approximation coefficients were appropriate for extracting the MMN from the averaged trace. The results obtained showed that wavelet decomposition (WLD) methods extract MMN as well as a band-pass digital filter (DF). The differences of the MMN peak latency between deviant types elicited by B-Spline WLD were more significant than those extracted by the DW, DF, or Reverse biorthogonal WLD. Also, wavelet coefficients of the delta-theta range indicated good discrimination between some combinations of the deviant types. У суб’єктів із нормальним слухом реєстрували пов’язані з подією потенціали, викликані з використанням множинної парадигми. Стандартні слухові стимули та девіантні стимули п’яти типів пред’являли в специфічній послідовності; ЕЕГ-потенціали відводили з частотою дискретизації 1024 c–1. Результати двох видів вейвлет-аналізу порівнювали з даними, отриманими із застосуванням традиційного методу диференціації хвиль (DW). Зворотний біортогональний вейвлет порядку 6.8 і квадратичний B-сплайновий вейвлет використовували для декомпозиції сьомого порядку. Коефіцієнти наближення шостого порядку виявилися застосовними для виділення негативності розузгодження (MMN) із усереднених записів. Як показали результати, методи вейвлет-декомпозиції (WLD) дозволяють виділити негативність розузгодження так само успішно, як і цифрові фільтри. Відмінності латентних періодів піків негативності розузгодження для девіантних варіантів стимуляції, виявлені в разі застосування В-сплайнової WLD, були більш вірогідними, ніж аналогічні відмінності при використанні методу диференціації хвиль, цифрової фільтрації або зворотної біортогональної WLD. Вейвлет-коефіцієнти для дельта-тета-діапазону також дозволяли отримати найкращу дискримінацію деяких комбінацій девіантних типів. 2014 Article Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm / M. Najafi-Koopaie, H. Sadjedi, S. Mahmoudian, E.D. Farahani, M. Mohebbi // Нейрофизиология. — 2014. — Т. 46, № 4. — С. 01-410. — Бібліогр.: 30 назв. — англ. 0028-2561 http://dspace.nbuv.gov.ua/handle/123456789/148296 612.014.42:519 en Нейрофизиология Інститут фізіології ім. О.О. Богомольця НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| description |
In this study, event-related potentials (ERPs) collected from normally hearing subjects and
elicited by a multi-feature paradigm were investigated, and mismatch negativity (MMN) was
detected. Standard stimuli and five types of deviant stimuli were presented in a specified
sequence, while EEG data were recorded digitally at a 1024 sec–1 sampling rate. Two wavelet
analyses were compared with a traditional difference-wave (DW) method. The Reverse
biorthogonal wavelet with an order of 6.8 and the quadratic B-Spline wavelet were applied
for seven-level decomposition. The sixth-level approximation coefficients were appropriate
for extracting the MMN from the averaged trace. The results obtained showed that wavelet
decomposition (WLD) methods extract MMN as well as a band-pass digital filter (DF). The
differences of the MMN peak latency between deviant types elicited by B-Spline WLD were
more significant than those extracted by the DW, DF, or Reverse biorthogonal WLD. Also,
wavelet coefficients of the delta-theta range indicated good discrimination between some
combinations of the deviant types. |
| format |
Article |
| author |
Najafi-Koopaie, M. Sadjedi, H. Mahmoudian, S. Farahani, E.D. Mohebbi, M. |
| spellingShingle |
Najafi-Koopaie, M. Sadjedi, H. Mahmoudian, S. Farahani, E.D. Mohebbi, M. Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm Нейрофизиология |
| author_facet |
Najafi-Koopaie, M. Sadjedi, H. Mahmoudian, S. Farahani, E.D. Mohebbi, M. |
| author_sort |
Najafi-Koopaie, M. |
| title |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm |
| title_short |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm |
| title_full |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm |
| title_fullStr |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm |
| title_full_unstemmed |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm |
| title_sort |
wavelet decomposition-based analysis of mismatch negativity elicited by a multi-feature paradigm |
| publisher |
Інститут фізіології ім. О.О. Богомольця НАН України |
| publishDate |
2014 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/148296 |
| citation_txt |
Wavelet Decomposition-Based Analysis of Mismatch Negativity Elicited by a Multi-Feature Paradigm / M. Najafi-Koopaie, H. Sadjedi, S. Mahmoudian, E.D. Farahani, M. Mohebbi // Нейрофизиология. — 2014. — Т. 46, № 4. — С. 01-410. — Бібліогр.: 30 назв. — англ. |
| series |
Нейрофизиология |
| work_keys_str_mv |
AT najafikoopaiem waveletdecompositionbasedanalysisofmismatchnegativityelicitedbyamultifeatureparadigm AT sadjedih waveletdecompositionbasedanalysisofmismatchnegativityelicitedbyamultifeatureparadigm AT mahmoudians waveletdecompositionbasedanalysisofmismatchnegativityelicitedbyamultifeatureparadigm AT farahanied waveletdecompositionbasedanalysisofmismatchnegativityelicitedbyamultifeatureparadigm AT mohebbim waveletdecompositionbasedanalysisofmismatchnegativityelicitedbyamultifeatureparadigm |
| first_indexed |
2025-07-12T19:04:52Z |
| last_indexed |
2025-07-12T19:04:52Z |
| _version_ |
1837469107711115264 |
| fulltext |
Article
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4 401
UDC 612.014.42:519
M. NAJAFI-KOOPAIE,1 H. SADJEDI,1 S. MAHMOUDIAN,2,3 E. D. FARAHANI,4 and M. MOHEBBI3
WAVELET DECOMPOSITION-BASED ANALYSIS OF MISMATCH NEGATIVITY
ELICITED BY A MULTI-FEATURE PARADIGM
Received 22.10.2013
In this study, event-related potentials (ERPs) collected from normally hearing subjects and
elicited by a multi-feature paradigm were investigated, and mismatch negativity (MMN) was
detected. Standard stimuli and five types of deviant stimuli were presented in a specified
sequence, while EEG data were recorded digitally at a 1024 sec–1 sampling rate. Two wavelet
analyses were compared with a traditional difference-wave (DW) method. The Reverse
biorthogonal wavelet with an order of 6.8 and the quadratic B-Spline wavelet were applied
for seven-level decomposition. The sixth-level approximation coefficients were appropriate
for extracting the MMN from the averaged trace. The results obtained showed that wavelet
decomposition (WLD) methods extract MMN as well as a band-pass digital filter (DF). The
differences of the MMN peak latency between deviant types elicited by B-Spline WLD were
more significant than those extracted by the DW, DF, or Reverse biorthogonal WLD. Also,
wavelet coefficients of the delta-theta range indicated good discrimination between some
combinations of the deviant types.
KEYWORDS: event-related potentials (ERPs), mismatch negativity (MMN), difference-
wave (DW), band-pass digital filter (DF), wavelet decomposition (WLD) techniques.
1 Electronics Group, Faculty of Engineering, Shahed University, Tehran, Iran
2 Department of Otorhinolaryngology, Hannover Medical University (MHH),
Hannover, Germany
3 ENT and Head and Neck Research Center, Tehran University of Medical
Sciences (TUMS), Tehran, Iran
4 Biomedical Engineering Faculty, Amirkabir University of Technology,
Tehran, Iran
Correspondence should be addressed to M. Najafi-Koopaie (e-mail:
mj.najafi@shahed.ac.ir, mojtabanajafi1366@gmail.com).
INTRODUCTION
Evoked potentials (EPs) or, more generally, event-
related potentials (ERPs) are defined as changes in the
electroencephalogram (EEG) related to certain events
(external stimulation or internal processes in the CNS).
Recently, mismatch negativity (MMN) studies of the
central auditory function have become very popular [1].
The MMN detection opened an unprecedented window
to the central auditory processing. The MMN, a change-
specific component of the auditory ERP, is elicited by
any discriminable change in auditory stimulation [2].
The MMN response is seen as a negative displacement
recorded, in particular from frontocentral and central
scalp sites relative to a mastoid or nose reference [3].
The new multi-feature paradigm was proposed by
Näätänen et al. [4] allowing one to obtain MMNs for
several auditory attributes within a short time. During
the experiment, standard stimuli and five different
types of deviant stimuli were presented. Figure 2
displays features of each stimulus in summary.
Duration ± 25 msec
Duration = 75 msec
Cutting out 7 msec
from the middle of the
standard stimulus,
leaving there
a gap
± 800 µsec latency
between two
channel (left/right)
sources
-
Equal phases at both ears
F i g. 2. Specifications of the standard and five types of deviant
stimuli in summary.
Р и с. 2. Стандартні стимули та девіантні стимули п’яти типів.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4402
M. NAJAFI-KOOPAIE, H. SADJEDI, S. MAHMOUDIAN et al.
Data extraction is crucially important in MMN
studies. Prevalently, after artifact rejection, recordings
belonging to each type of stimulus are averaged to
obtain the ERP waveform. Responses to standard
stimuli are typically subtracted from the ERPs
elicited by infrequently presented deviant stimuli. The
resulting wave is called the difference wave (DW),
which indicates the MMN [5]. The peak amplitude and
peak latency of MMN are usually measured from the
DW. This is the most typical processing used for MMN
detection.
Generally, the signal processing techniques to extract
MMN are divided into two categories, single-channel
and multi-channel procedures. Some methods, such as
difference-wave (DW), digital filters (DFs) [6, 7], and
wavelet decomposition (WLD [8] can be applied to one
channel, and these are the single-channel procedures.
Other methods, such as component analysis [9, 10] and
matrix factorization [11], need more than one channel
and are categorized as multi-channel procedures.
The idea of using the adaptive filtering technique for
the analyzing of EPs was first proposed by Orfanidis [12]
and Thakor [13] and, later on, was investigated by other
authors [14-16]. Researchers employed time-frequency
analysis to gain understanding of mass electrical activity
of the brain. The wavelet transform has been applied to a
few bioelectric signals. Thakor et al. [17] used a wavelet-
based method for the analysis of ECG data, and Schiff et
al. [18] used this approach for EEG.
Wavelet filters were especially designed for non-
stationary signals; they utilize both time and frequency
information related to a signal. WLD techniques
factorize the signal into several levels with a particular
wavelet at first, and then coefficients of the selected
levels can be used for reconstruction or comparison
[19]. Each level of decomposition matches to a
certain frequency band, although frequency bands of
neighboring levels may overlap each other around the
cut-off frequencies. Since the selected wavelet for
decomposition and reconstruction may be correlated
with the desired signal, overlapping signals can be
separated by a DF.
In our study, five types of MMN were obtained
using five different types of deviant stimuli presented
in a special sequence. EEG signals of young people
were recorded, and MMNs were analyzed with
quadratic B-Spline WLD. To validate the effectiveness
of the proposed methods, the results were compared
with the average-based DW (calculated by subtracting
the standard response from the deviant response), a DF
technique, and a Reverse biorthogonal WLD with an
order of 6.8 [20].
METHODS
Subjects. The group of participants consisted of
43 healthy normally hearing volunteers, 21 men and
22 women. They were between 20 to 24 years old,
with no history of auditory disorders, and most of
them were university students.
Stimuli. In this study, MMNs were obtained using
the new paradigm proposed by Näätänen et al. [4],
with a little change in the number of stimuli to shrink
the time of recording. This paradigm makes it possible
to obtain five types of MMN in a considerably shorter
recording time compared with the traditional oddball
conditions. In the new paradigm, each deviant is
presented after a standard stimulus, meaning that the
deviants occur with the probability of 0.5 relative
to the standards (PStd=0.5, PDev=0.5/5=0.1). Standard
stimuli were composed of three sinusoidal tones
of 500, 1000, and 1500 Hz with a total duration of
75 msec including 5-msec rise and 5-msec fall time.
The intensity of the second and third tones sequentially
was 3 and 6 dB lower than the first tone, respectively.
The stimuli were presented binaurally via ER-3A
insert earphones with an intensity level of 65 dB SPL
and equal phase in both ears.
The deviant stimuli were generated differently from
the standards in five categories. These differences were
in the frequency, intensity, duration, perceived sound-
source location, and a gap in the middle of the tone sig-
nal. Frequency, intensity, and location deviants were in
two modes. A half of the frequency deviant tones were
10% higher (550, 1100, 1650 Hz), while another half
were 10% lower (450, 900, 1350 Hz) than the standard.
A half of the intensity deviants were 10 dB lower and
another half were 10 dB higher than the standard. De-
viants of location were generated based on a change
in the location of a perceived sound source. An inter-
aural time difference of 800 µsec was applied for half
of the deviants to the right channel and another half to
the left channel. The duration deviant was shorter than
in duration (5-msec rise, 15-msec flat, and 5-msec fall
times). The silent gap deviant consisted of a 7-msec si-
lent gap (including 1-msec rise and 1-msec fall times)
in the middle of the standard stimulus. Fig. 1 shows the
waveforms of a standard stimulus, duration deviant, and
silent gap deviant. Each deviant differed from the stan-
dard only in one feature. Features of the standard and
deviants stimuli are shown in Fig. 2 in summary.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4 403
WAVELET DECOMPOSITION-BASED ANALYSIS OF MISMATCH NEGATIVITY ELICITED BY A MULTI-FEATURE PARADIGM
back were supported with pillows to reduce muscle
contractions. They, after setting an EEG cap on the
scalp, were instructed to be relaxed, ignore the
auditory stimuli, and stay awake. A subtitled silent
movie was played on a front monitor, to maintain
alertness and to help participants to pay no attention to
the stimuli during the experiment. The EEG recording
session, including preparation and recording per se,
lasted about 30 min.
EEG Recording. Sixty-four-channel BRAIN
QUICK LTM (Micromed, Italy) was used for
recording electrical brain activities. Twenty-seven
EEG channels were used. Ag-AgCl electrodes were
filled with Electro-Gel and placed on 27 selected
scalp sites (FP1, FPz, FP2, F7, F3, Fz, F4, F8, FT7,
FC3, FCz, FC4, FT8, T7, C3, Cz, C4, T8, TP7,
CP3, CPz, CP4, TP8, P3, Pz, P4, and POz) and
mastoids (M1 and M2), according to the international
10-20 system. The potentials were referred to the nose
tip. Electrooculogram activity (EOG) was recorded by
placing electrodes below the left eye and at its outer
canthus. Impedances during recording did not exceed
5 kΩ. EEG signals were digitized with the sampling
rate of 1024 sec–1 and filtered by an online band-pass
filter in the 0.001-100 Hz range. In addition, a custom-
designed microcontroller device received digital
events from Neurobehavioral presentation software
and reformed it to compatible trigger signals to mark
events on the computerized EEG record.
EEG Data Preprocessing. The EEG data were
analyzed off-line using MATLAB® 7.1. At first, these
data were filtered by an off-line band-pass digital filter
in the 0.5-40 Hz range. Then, epochs were extracted
from the continuous data according to a trigger signal
0
0
0 10 15 20 25 30
10 20 30 40 50 60 70 75
msec
msec
–0.5
0.5
A
B
C
0
–0.5
0.5
0
–0.5
0.5
0 10 20 30 40 50 60 70 75
msec
F i g. 1. Waveforms of a standard stimulus (A), duration deviant (B),
and silent-gap deviant (C) generated by MATLAB software®.
Р и с. 1. Форми стандартного стимулу (А), девіанта щодо три-
ва лості (В) та девіанта з „вікном мовчання” (С), генерованих з
використанням MATLAB.
The stimuli were presented in two 5-min-long
blocks with a 500 msec onset asynchrony. In each
block, the first 15 stimulus were standards, and the
deviants were presented pseudo-randomly within the
stimuli. So in an array of five deviants, each deviant
was presented once, and two similar types of deviants
never followed each other (Fig. 3). A total of 1,230
stimuli was presented within the total recording time
(about 10 min) for the five types of deviants. Stimuli
were presented via Presentation® software (version
0.71, Neurobehavioral Systems©, USA). This software
is a specialized stimulus delivery and experimental
control program for neuroscientific research purposes.
Procedure. EEG was recorded in electromagnetic-
and sound-proof chamber. Participants were seated
on a comfortable chair, and their head, neck, and
P
P
500 msec
F i g. 3. Sequence of presentation of the stimuli. S indicates a
standard stimulus, and Dx indicates different deviant types.
Р и с. 3. Послідовність пред’явлення стимулів.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4404
M. NAJAFI-KOOPAIE, H. SADJEDI, S. MAHMOUDIAN et al.
that was generated by microcontroller device. All data
were baseline-corrected by 100 msec pre-stimulus.
The EEG data were checked for blink, ECG, and other
muscular artifacts by visual inspection. Epochs with
artifacts were rejected from subsequent processing.
In addition, epochs where the amplitude exceed
80 µV were automatically rejected. The mean number
of trials (after artifact rejection) per subject was 1027.
Finally, epochs of the standards and all type of deviants
were averaged within 100 msec pre- stimulus to 380 msec
post-stimulus segments separately. The first 15
standards of each block were rejected from averaging.
EEG data analysis. In several researches, various
frequency bands were used to filter out interference from
an averaged MMN trace. Kalyakin et al. [21] reported
that the optimum frequency band of MMN of children
was 2.0–8.8 Hz for the uninterrupted sound paradigm;
Stefanics et al. [7] used two frequency bands (2.5–16 and
1.5–16 Hz) to obtain the MMN of neonates. Picton et al.
[22] reported that most MMN energy was concentrated
within the 2–5 Hz frequency range. Tervaniemi et al. [23]
used a 2–12 Hz band-pass digital filter for analyzing the
peak amplitude and latency of MMN.
The wavelet transform gives a time frequency
representation of a signal that is defined as the
convolution between the signal x(t) and the wavelet
function ѱa,b(t)
,
where ѱa,b(t) are dilated and shifted versions of a
unique wavelet function ѱ(t)
.
Here, a and b are the scale and translation
parameters, respectively [24]. Discrete wavelet
transform (DWT) applies to discrete time signals x[n].
It achieves a multiresolution decomposition of x[n] on
I octaves labeled by i = 1, ... , I and given by
.
The DWT calculates the wavelet coefficients ai,k for
i = 1, ... , I and the scaling coefficients bi,k. The latter
are given by
and
,
where g i[n – 2 ik]s are the discrete wavelets, and
hI[n – 2Ik]s are the scaling sequences [25].
A basic wavelet function to be compared with the
signal should be chosen. There are many different
functions suitable as wavelet, each one having
different characteristics. One hundred ten wavelets
consisting of different orders of Daubechies, Coiflets,
Symlets, discrete Meyer, biorthogonal, and reverse
biorthogoanl wavelets were compared by Cong et al.
[20]. Finally, the Reverse biorthogonal wavelet with an
order of 6.8 was chosen for wavelet decomposition of
MMN. The Spline wavelet was used to study the P300
of young people by Demiralp et al. [26]. Ademoglu et
al. also used quadratic spline wavelet to characterize
the N70-P100-N130 EP complex [27].
Four data processing methods, DW, DF, and WLD
with two different-type wavelet methods were used
for comparison of the results. The MMN response is
typically obtained in frontocentral sites better than in
others [28]; so, the following processes were applied
in site FCz. In each method, MMN properties (peak
amplitude and latency) were extracted.
For DW, traces were calculated by subtracting the
responses to the standard stimuli from each type of
deviants. The peak amplitude and latency were calculated
in FCz for each subject. These properties were obtained
based on the highest negativity of averaged MMN within
a time window of 100–230 msec post-stimulus.
The DF was applied in three steps. Fourier transform
of the signals was performed on average traces; then,
Fourier confidents outside the 1–8 Hz range were set
to zero, and, finally, inverse Fourier transform was
used to obtain the filtered traces.
For WLD, two different wavelets were used
to decompose the signal into seven levels, and
approximation coefficients of the sixth level were
selected for reconstruction. Filter coefficients
corresponding to quadratic B-Spline wavelet were
computed as was described by Ademoglu et al. [27].
Filter coefficients for Reverse biorthogonal were
obtained by MATLAB. WLD with quadratic B-Spline
wavelet is denoted below as WLD-BS, and WLD by
Reverse biorthogonal with an order of 6.8 is denoted
as WLD-RB.
In WLD, if the number of the decomposition levels
is L, the number of samples of the signal per one
second is N under conditions of N = 2L [6].
In our study, the sampling frequency for EEG
data recording was set to 1024 sec–1, so, the signal
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4 405
WAVELET DECOMPOSITION-BASED ANALYSIS OF MISMATCH NEGATIVITY ELICITED BY A MULTI-FEATURE PARADIGM
could be composed of ten levels. The bandwidths at
each decomposition level are shown in Table 1. The
frequencies are estimates of the bandwidth of each
level, and these frequencies are not related to the
properties of any wavelet.
Because the off-line digital band-pass filter was
applied in 0.5-40 Hz, there is no useful data within
the frequency band for D1 to D4. For WLD, the data
were decomposed into seven levels. As is indicated
in Table 1, the frequency range for A6 matched best
the optimum frequency range of MMN. This frequency
band corresponds to a delta-theta range of EEG
signals.
The DW, DF, and wavelet filters have the linear
additive property. Hence, first averaged traces were
calculated; then WLD and DF were applied to reduce
the computation loading.
RESULTS
Analysis of the efficacy of main effect of MMN
measurements based on different methods was the
main purpose of our study. The first MMN peak
amplitude and latency were detected in each MMN for
each subject, and then these data were examined using
repeated-measures analysis of variance (ANOVA) to
determine whether a difference of MMN properties
(peak amplitude and latency) between five deviants
was evident under each method used, respectively.
Also, we wanted to determine whether the differences
of MMN properties between WLD-BS and the DW,
DF, or WLD-RB are significant.
Figure 4 shows grand averaged waveforms obtained
using DW, DF, and WLD procedures for each type of
MMN. The thick solid, thin solid, dashed, and dotted
0
0.5
–0.5
–1.0
–1.5
–2.0
–2.5
–100 50 100 150 200 250 300 350 msec–50 0
–100 50 100 150 200 250 300 350 msec–50 0
1.0
1.5
mV
0
0.5
–0.5
–1.0
–1.5
–2.0
–2.5
1.0
1.5
0
0.5
–0.5
–1.0
–1.5
–2.0
–2.5
1.0
1.5
F i g. 4. Grand-averaged traces recorded from site FCz for
difference-wave (DW, 1), digital filter (DF, 2), B-Spline wavelet
(WLD-BS, 3), and Reverse biorthogonal wavelet (WLD-RB, 4) of
all deviant responses (A-E).
Р и с. 4. Усереднені записи негативності розузгодження,
відведеної від локусу FCz та виділеної з використанням
диференціації хвиль (DW, 1), цифрової фільтрації (DF, 2),
B-сплайн-вейвлет-декомпозиції (WLD-BS, 3) та зворотної
біортогональної вейвлет-декомпозиції (WLD-RB, 4), для всіх
девіантних відповідей (A-E).
A
Frequency
1
2
3
4
3
4
2
1
Duration
Silent-gap
Intensity
Location
C
E
B
D
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4406
M. NAJAFI-KOOPAIE, H. SADJEDI, S. MAHMOUDIAN et al.
lines represent the WLD-BS, WLD-RB, DF, and DW
traces, respectively. A gray field designates the time
window where the MMN peak amplitude was detected.
As is shown, MMN did not appear for the intensity
deviant in a same wave as some other researchers
reported [29]; this is why we excluded it from the
subsequent analysis.
The MMN peak latencies detected by different
methods were found to be very similar (F4,42 = 0.96,
P = 0.41), although this was not true for the MMN
peak amplitudes (F4,42 = 38.02, P < 0.000). The peak
amplitudes obtained by DW differed considerably
from those obtained by WLDs or DF.
To investigate which MMN extraction method is
better, the abilities of these methods to discriminate
between different MMN types were compared. Hence,
the MMN properties (peak latency and amplitude)
extracted by the above processing methods were
examined in all combinations.
The results of MMN extraction between each type
of deviants using different methods are shown in
Fig. 5. All statistical tests of the differences between
the MMN peak amplitudes and latencies elicited by
four deviants using four methods are collected in
this Figure. Two horizontal lines indicate 0.05 and
0.01 borders for the P value to determine whether a
result is statistically significant. For all methods, the
differences of the peak magnitude or latency between
location and silent-gap deviants were not significant.
The peak latency differences between frequency and
duration deviants were also not significant for all
methods, and the difference of the peak amplitude
between these two deviants was significant for DW.
Main effects of all methods for the peak latency in all
combinations of deviants were similar to each other, as
F i g. 5. Statistical tests of the differences between four types of the
MMN amplitude and latency extracted by WLD-BS (1), WLD-RB
(2), DF (3), and DW (4).
Р и с. 5. Статистичні тести щодо різниць між негативностями
розузгодження за амплітудою (А) та латентним періодом (В),
виділених з використанням WLD-BS (1), WLD-RB (2), DF (3)
та DW (4).
T a b l e 1. Frequency Levels for Wavelet Decomposition
Т а б л и ц я 1. Частотні рівні для вейвлет-декомпозиції
Decomposition level Approximation and detail coeff. Frequency range (Hz) Bandwidth (Hz)
7 A7 0–4 4D7 4–8
6 A6 0–8 8D6 8–16
5 A5 0–16 16D5 16–32
4 A4 0–32 32D4 32–64
3 A3 0–64 64D3 64–128
2 A2 0–128 128D2 128–256
1
A1 0–256
256D1 256–512
1
1
10–1
10–1
10–2
10–3
10–4
Frequency
Duratio
n
Frequency
Loca
tio
n
Freq
ue
nc
y
Sile
nt
Gap
Dura
tio
n
Lo
ca
tio
n
Dura
tio
n
Sile
nt
Gap
Lo
ca
tio
n
Sile
nt
Gap
10–2
10–3
2
2
4
4
3
3
1
1
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4 407
WAVELET DECOMPOSITION-BASED ANALYSIS OF MISMATCH NEGATIVITY ELICITED BY A MULTI-FEATURE PARADIGM
it was expected from averaged traces in Fig. 4. Totally,
the peak latency in WLD-BS for all combinations of
deviants was significantly greater than that in other
methods.
In wavelet-based MMN extraction methods,
coefficients of various decomposition levels consist
of time-frequency information, in contrast to DW or
DF that only have data of either time or frequency
domains. According to the frequency band for
MMN (about 1 to 10 Hz), the coefficients of sixth-
level decomposition were compared together. Figure
6 shows statistical comparison of the test results of
different types of responses for WLD methods. The
WLD coefficients of averaged traces were compared
by WLD-BS and WLD-RB; in column A, this is
performed for standard and four deviants and in
column B, this is performed for different waves. As it
shown, discriminations are similar to each other and
independent of the standard, i.e., whether a standard
sweep was subtracted from the deviant sweeps or
not. It should be noted that critical values from the t
distribution were used after Bonferroni adjustment, to
compensate multiple comparisons.
The WLD coefficients of delta-theta range for the
standards were significantly greater than those for the
deviants. There are main effects with respect to the
deviant pairs, frequency/location, frequency/silent
gap, duration/location, and duration/silent gap. These
results are obvious in Fig .5 if the latency was the
comparison factor. There was no main effect between
the location and silent-gap deviants for each method,
either for the peak latency and amplitude, or for the
WLD coefficients.
DISCUSSION
In our study, the criteria used for evaluating the
performance of the data processing methods were
based on the MMN properties, i.e., it was believed
that different types elicit different MMNs [30]. The
experiment included five deviants differing in the
frequency, intensity, duration, perceived location,
and silent gap. The WLDs gave the actual MMN peak
magnitude and latency, as was confirmed by analyzing
the MMN properties between deviants (see Fig. 5).
Table 2 shows statistical test results on the MMN
peak amplitude and latency between WLD-BS and
the DF, DW, or WLD-RB. For ANOVA, the method
was the factor. The respective results show that the
proposed WLD-BS performed differently with the DW
in extracting MMN. However, there is a main effect
between these two methods in extracting the MMN
peak amplitude; they provided similar discriminations
between different deviants (see Fig. 5).
WLD can be regarded as a special bandpass filter.
The frequency responses of quadratic B-Spline WLD
and Reverse biorthogonal WLD with the order of
6.8 are shown in Fig. 7, and their filter coefficients are
shown in Fig. 8. The wavelet morphologies are similar,
while the frequency responses are different. Reverse
boirthogonal 6.8 is alike to be an ideal bandpass
Frequency
Devi ant
Duration
Devi ant
Location
Devi ant
Silent Gap
Devi ant
Standard
Frequency
Devi ant
Duration
Devi ant
Location
Devi ant
Silent Gap
Devi ant
Standard
–1.3
–10 –8 –6 –4 –2 0 2 4 6 8 –10 –8 –6 –4 –2 0 2 4 6 8
–1.0 –0.5 0.5 1.01.20 –1.3–1.0 –0.5 0.5 1.01.20
F i g. 6. Comparison tests (with Bonferroni
adjustment) between MMN delta-theta-range
coefficients extracted by WLD-BS and WLD-
RB. The WLD coefficients for averaged deviant
responses and standards are shown in column
A); WLD coefficients for difference waves are
shown in column B.
Р и с. 6. Тести порівняння (з наближен ням
Бонферроні) між дельта-тета-коефіцієнтами
негативності розузгодження, виділеними з
використанням WLD-BS та WLD-RB.
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4408
M. NAJAFI-KOOPAIE, H. SADJEDI, S. MAHMOUDIAN et al.
filter in contrast to quadratic B-Spline that amplifies
frequencies close to the cut-off frequency.
Considering frequency information of MMN, we
need the filter having a better frequency match with
the MMN frequency range to extract pure MMN.
It seems that the WLD-BS is a good method for
extracting MMN with better properties.
In our study, the compared results showed that
there is no disparity between the WLD coefficients of
deviants and WLD coefficients of DW (see Fig. 6).
Hence, we can apply WLD to deviants directly instead
of DW.
The WLD approach was applied for investigation
of the differences between types of MMN obtained by
the multi-feature paradigm in a sampling of normally
hearing people. From this point of view, the WLD
method and WLD coefficients can be used with
respect to other subjects, e.g., complainers of hearing
disorders; these also can be used to specify some brain
pathologies.
Ideally, only MMN activity should remain in the
data for detecting properties or feature extraction after
data processing. However, the DW only removes the
common variance in standard and deviants traces;
other types of activity that overlap MMN are not
segregated just in the time or frequency domain. Thus,
time-frequency processing can be used to obtain pure
MMNs; time and frequency information should be
applied together for analyzing. This matter motivated
us to use time-frequency analyzing based on WLD and
to compare types of MMN elicited by the specified
new paradigm. With this approach, we have found
that the WLD coefficients are better tools to compare
MMNs than estimation of traditional MMN properties
(peak latency and amplitude).
The Ethics Committee of ENT and Head and Neck Research
Center, Tehran University of Medical Sciences, acknowledged
the study design (code number: MT.8829/90-12-25) as
corresponding to the internationally accepted ethic standards.
All participants of the tests were informed in detail on the
0
–0.2
–0.5
0
0.5
A
B
0.4
0.8
0
5 10 15 20
F i g. 8. Filter coefficients for quadratic B-Spline wavelet (A) and
Reverse biorthogonal 6.8 wavelet (B).
Р и с. 8. Коефіцієнти фільтрації для квадратичного В-сплайн-
вейлвета (А) та зворотного біортогонального 6.8 вейлвета (В).
0
F/ 2 F
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
F i g. 7. Frequency characteristics of the wavelet filters (1 and 2, for
B-Spline and Reverse biorthogonal 6.8, respectively).
Р и с. 7. Частотні характеристики вейвлет-фільтрів (1 – для
В-сплай нового, 2 – для зворотного біортогонального).
Table 2. Statistical Tests of the Differences Between WLD-BS and Other Methods in the Analysis of the Peak Amplitude and Latency
Т а б л и ц я 2. Статистичні тести щодо різниць між WLD-BS та іншими методами при аналізі пікових амплітуд та латентних
періодів
Parameter Value
Methods
WLD-BS vs DW WLD-BS vs DF WLD-BS vs WLD-RB
Amplitude
F(1,42) 83.79 3.85 0.83
P <0.0000 0.0504 0.3621
Latency
F(1,42) 3.64 0.43 0.52
P 0.0574 0.5141 0.4693
1
2
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4 409
WAVELET DECOMPOSITION-BASED ANALYSIS OF MISMATCH NEGATIVITY ELICITED BY A MULTI-FEATURE PARADIGM
experimental procedure and gave their written consent.
The au thors , M. Na ja f i -Koopa ie , H . Sad jed i ,
S. Mahmoudian, E. D. Farahani, and M. Mohebbi, confirm that
they have no conflict of interest.
М. Наджафі-Купайє1, Х. Саджеді1, С. Махмудіан2,3,
Е. Д. Фарахані4, М. Мохеббі3
ЕЕГ-НЕГАТИВНІСТЬ РОЗУЗГОДЖЕННЯ,
ЗАРЕЄСТРОВАНА В УМОВАХ МНОЖИННОЇ ПА-
РАДИГМИ: АНАЛІЗ, ЗАСНОВАНИЙ НА ВЕЙВЛЕТ-
ДЕКОМПОЗИЦІЇ
1 Університет Шахед, Тегеран (Іран).
2 Ганноверський медичний університет (ФРН).
3 Тегеранський університет медичних наук (Іран).
4 Технологічний університет Аміркабір, Тегеран (Іран).
Р е з ю м е
У суб’єктів із нормальним слухом реєстрували пов’язані
з подією потенціали, викликані з використанням множин-
ної парадигми. Стандартні слухові стимули та девіантні
стимули п’яти типів пред’являли в специфічній послідов-
ності; ЕЕГ-потенціали відводили з частотою дискретизації
1024 c–1. Результати двох видів вейвлет-аналізу порівнювали
з даними, отриманими із застосуванням традиційного мето-
ду диференціації хвиль (DW). Зворотний біортогональний
вейвлет порядку 6.8 і квадратичний B-сплайновий вейвлет
використовували для декомпозиції сьомого порядку. Коефі-
цієнти наближення шостого порядку виявилися застосов-
ними для виділення негативності розузгодження (MMN) із
усереднених записів. Як показали результати, методи вейв-
лет-декомпозиції (WLD) дозволяють виділити негативність
розузгодження так само успішно, як і цифрові фільтри. Від-
мінності латентних періодів піків негативності розузго-
дження для девіантних варіантів стимуляції, виявлені в разі
застосування В-сплайнової WLD, були більш вірогідними,
ніж аналогічні відмінності при використанні методу дифе-
ренціації хвиль, цифрової фільтрації або зворотної біорто-
гональної WLD. Вейвлет-коефіцієнти для дельта-тета-діа-
пазону також дозволяли отримати найкращу дискримінацію
деяких комбінацій девіантних типів.
REFERENCES
1. R. Näätänen, A. W. K. Gillard, and S. Mäntysalo, “Early
selective-attention effect on evoked potential reinterpreted,”
Acta Psychol. (Amst.), 42, 313-329 (1978).
2. R. Näätänen and I. Winkler, “The concept of auditory stimulus
representation in cognitive neuroscience,” Psychol. Bull., 125,
826–859 (1999).
3. R. Näätänen, P. Paavilainen, T. Rinne, and K. Alho, “The
mismatch negativity (MMN) in basic research of central
auditory processing: A review,” Clin. Neurophysiol., 118,
2544-2590 (2007).
4. R. Näätänen, S. Pakarinen, T. Rinne, and R. Takegata, “The
mismatch negativity (MMN): towards the optimal paradigm,”
Clin. Neurophysiol., 115, 140-144 (2004).
5. R. Näätänen, Attention and Brain Functions, Hillsdale, New
Jersey (1992).
6. J. Sinkkonen and M. Tervaniemi, “Towards optimal recording
and analysis of the mismatch negativity,” Audiol. Neurootol.,
5, 235-246 (2000).
7. G. Stefanics, G. Haden, L. Balazs, et al., “Auditory temporal
grouping in newborn infants,” Psychophysiology, 44, 697-702
(2007).
8. M. Burgur, U. Hoppe, P. Kummer, et al., “Wavelet-based
analysis of MMN responses in children,” Biomed. Technol.,
52, 11-116 (2007).
9. F. Cong, I. Kalyakin, H. Li, et al., “Answering six questions in
extracting children’s mismatch negativity through combining
wavelet decomposition and independent component analysis,”
Cogn. Neurodyn., 5, 343-359 (2011).
10. J. Dien, “The ERP PCA toolkit: an open source program for
advanced statistical analysis of event related potential data,”
J. Neurosci. Methods, 187, 138-145 (2010).
11. F. Cong, I. Kalyakin, A. Phan, et al., “Identical fits of
nonnegative matrix/tensor factorization may correspond to
different extracted event-related potentials,” Proc. Int. Jt.
Conf. Neural Netw., 2260-2264 (2010).
12. S. Orfanadis, F. Afif, and E. Micheli-Tzanakou, “Visual evoked
potentials extraction by adaptive filtering,” Conf. Proc. IEEE
Eng. Med. Biol. Soc., 968-969 (1987).
13. N. Thakor, “Adapting filtering of evoked potentials,” IEEE
Trans. Biomed. Eng., BME-33, 1085-1095 (1987).
14. E. Hamzaoui, F. Regragui, M. M. Himmi, and A. Mghari,
“Visual evoked potentials’ non linear adaptive filtering based
on three layers perceptron,” Appl. Math. Sci. (Ruse), 4, 2759-
2772 (2010).
15. P. Laguna, R. Jane, O. Meste, et al., “Adaptive filter for
event-related bioelectric signals using an impulse correlated
reference input,” IEEE Trans. Biomed. Eng., 39, 1032-1044
(1992).
16. C. A. Vaz and N. V. Thakor, “Adaptive Fourier estimation of
time-varying evoked potentials,” IEEE Trans. Biomed. Eng.,
36, 448-455 (1989).
17. N. V. Thakor, Y. Sun, H. Rix, and P. Caminal, “Multiwave:
a wavelet-based ECG data compression algorithm,” IEICE
Trans. Inf. Syst., E-76D, 1462-1469 (1993).
18. S. J. Schif, A. Aldroubi, M. Unser, and S. Sato, “Fast
wavelet transformation of EEG,” Electroencephalogr. Clin.
Neurophysiol., 91, 442-455 (1994).
19. V. J. Samar, K.P. Swartz, and M.R. Raghuveer, “Multiresolution
analysis of event-related potentials by wavelet decomposition,”
Brain Cogn., 27, 398-438 (1995).
20. F. Cong, Y. Huang, I. Kalyakin, et al., “Frequency-response-
based wavelet decomposition for extracting children’s
mismatch negativity elicited by uninterrupted sound,” Med.
Biol. Eng., 32, 205-214 (2012).
21. I. Kalyakin, N. Gonzalez, J.Joutsensalo, et al., “Optimal digital
filtering versus difference wave on the mismatch negativity in
an interrupted sound paradigm,” Dev. Neuropsychol., 31, 429-
452 (2007).
NEUROPHYSIOLOGY / НЕЙРОФИЗИОЛОГИЯ.—2014.—T. 46, № 4410
M. NAJAFI-KOOPAIE, H. SADJEDI, S. MAHMOUDIAN et al.
22. T. Picton, C. Alain, L. Otten, et al., “Mismatch negativity:
different water in the same river,” Audiol. Neurootol., 5, 111-
139 (2000).
23. M. Tervaniemi, A. Lehtokoski, J. Sinkkonen, et al., “Test-retest
reliability of mismatch negativity for duration, frequency
and intensity changes,” Clin. Neurophysiol., 110, 1388-1393
(1999).
24. A. Grossman and J. Morlet, “Decomposition of hardy
functions into square integrable wavelets of constant shape,”
SIAM J. Math. Anal., 15, 723-736 (1984).
25. S. Mallat, “A theory for multiresolution signal decomposition:
the wavelet representation,” IEEE Trans. Pattern. Anal. Mach.
Intell., 11, 674-693 (1989).
26. T. Demiralp, Y. Istefanopulos, A. Ademoglu, et al., “Analysis
of functional components of P300 by wavelet transform,”
Conf. Proc. IEEE Eng. Med. Biol. Soc., 20, 1992-1995 (1998).
27. A. Ademoglu, E. Micheli-Tzanakou, and Y. Istefanopulos,
“Analysis of pattern reversal visual evoked potential
(PRVEP’s) by spline wavelet,” IEEE Trans. Biomed. Eng., 44,
881-889 (1997).
28. T. Kujala, M. Tervaniemi, and E. Schröger, “The mismatch
negativity in cognitive and clinical neuroscience: Theoretical
and methodological considerations,” Biol. Psychol., 74, 1-19
(2007).
29. H. Thönnessen, M. Zvyagintsev, K. C. Harke, et al.,
“Optimized mismatch negativity paradigm reflects deficits in
schizophrenia patients a combined EEG and MEG study,” Biol.
Psychol., 77, 205-216 (2008).
30. R. Näätänen, S. Pakarinen, T. Rinne, and R. Takegata, “The
mismatch negativity (MMN): towards the optimal paradigm,”
Clin. Neurophysiol., 115, 140-144 (2004).
|