J Med Syst
DOI 10.1007/s10916-010-9498-8
ORIGINAL PAPER
Detection of Carotid Artery Disease by Using Learning
Vector Quantization Neural Network
Harun Uğuz
Received: 30 January 2010 / Accepted: 12 April 2010
# Springer Science+Business Media, LLC 2010
Abstract Doppler ultrasound has been usually preferred
for investigation of the artery conditions in the last two
decades, because it is a non-invasive, easy to apply and
reliable technique. In this study, a biomedical system based
on Learning Vector Quantization Neural Network (LVQ
NN) has been developed in order to classify the internal
carotid artery Doppler signals obtained from the 191
subjects, 136 of them had suffered from internal carotid
artery stenosis and rest of them had been healthy subject. The
system is composed of feature extraction and classification
parts, basically. In the feature extraction stage, power spectral
density (PSD) estimates of internal carotid artery Doppler
signals were obtained by using Burg autoregressive (AR)
spectrum analysis technique in order to obtain medical
information. In the classification stage, LVQ NN was used
classify features from Burg AR method. In experiments, LVQ
NN based method reached 97.91% classification accuracy
with 5 fold Cross Validation (CV) technique. In addition, the
classification performance of the LVQ NN was compared with
some methods such as Multi Layer Perceptron (MLP) NN,
Naive Bayes (NB), K-Nearest Neighbor (KNN), decision tree
and Support Vector Machine (SVM) with sensitivity and
specificity statistical parameters. The classification results
showed that the LVQ NN method is effective for
classification of internal carotid artery Doppler signals.
Keywords Learning vector quantization . Doppler signal .
Carotid artery . Power spectral density . Autoregressive
method
H. Uğuz (*)
Department of Computer Engineering, Selçuk University,
Konya, Turkey
e-mail: harun_uguz@selcuk.edu.tr
e-mail: harun_uguz@hotmail.com
Introduction
Carotid artery is a disease negatively affecting the vessels
going to the head and brain. The symptoms of carotid artery
disease are stenosis and occlusions occurring in the internal
carotid artery. These symptoms are led by the internal
carotid artery plaques [3]. Angiography and blood tests are
the methods which can be referred to for diagnostics of the
disease. On the other hand, these techniques are not
preferred since they are invasive.
Doppler technique was used as first time in 1959 by
Satomura. Since this technique is noninvasive and has no
risk [4], it is often referred to in demonstrating the flow
characters of carotid arteries. As a medical application of
Doppler principle, an ultrasonic wave is sent to the body and
the shift of the wave’s frequency is measured detecting the
reflected wave from the red blood cells in the body. The
Doppler signal frequency difference between the incoming
and reflected waves is directly proportional to the blood
velocity. The scattering caused by the reflected incoming
waves from the red blood cells exhibit a frequency spectrum.
Conclusions on blood flow can be revealed by examining the
changes within this spectrum. On the basis of the spectral
form and parameter analysis, it can be determined whether
symptom belongs to a healthy or diseased artery [5, 6].
PSD estimates can be used to track the velocity
distribution, since the velocity components will be propor-
tional to the frequency shifts. Spectrum analysis techniques
were used to obtain the Doppler PSD estimates of carotid
arterial Doppler signals. Recently, making use of various
spectrum analysis methods in the processing of internal
carotid artery Doppler signals [7–10]. AR method is a
common practice based on the fact that solving linear
equations makes it easy to estimate the AR parameters. For
that reason, AR method is widely used as a spectrum

analysis technique [6, 8–13]. Yule-Walker, Burg, covari-
ance, least squares, and maximum likelihood estimation are
examples to the various estimation methods for the
parameters of AR method [1, 2]. Burg AR method stands
as a powerful calculation method in terms of computational
efficiency, and resulting in consistent estimates, is frequent-
ly used in PSD estimates of internal carotid artery Doppler
signals [8, 19, 20].
In the literature, there are various studies on classifica-
tion of the carotid artery Doppler signals. Özşen et al.
extracted the features with AR method and classified these
features using a new Artificial Immune Systems classifier
[11]. Ceylan et al. used Complex Valued Artificial Neural
Network (CVANN) structure to classify carotid artery
Doppler signals using Principal Component Analysis and
Fuzzy-C-Means Clustering (FCM) as feature extraction
methods before the CVANN classifier [14]. Özbay &
Ceylan used Fast Fourier Transform, Hilbert Transform,
and Welch Method with different window types. They
investigated effects of window types on classification of
carotid artery Doppler signals [15]. Polat et al. used Support
Vector Machine (LSSVM) with a fuzzy-weighted prepro-
cessing step in classification progress [16]. In Güler &
Übeyli’s studies, attributes attained from the carotid artery
Doppler signals by the help of the various spectral analysis
methods were classified according to MLP NN-based
classification systems [8, 17, 18].
Artificial Neural Network (ANN) is a calculation system,
having been developed through inspiration from the structure,
as well as from the learning characteristics of neural cells. It is
similarly and successfully applied on the functional features of
human brain, in terms of such aspects as learning, optimiza-
tion, prediction, clustering, generalization, and classification.
Main reasons of the frequent use of ANN in classification
applications are as follows [21]: (i) ANN’s simple structure for
easier use in hardware platform; (ii) ANN’s easier mapping of
complex class-distributed features; (iii) ANN’s generalization
feature. Its generation of results appropriate for input vectors
unavailable in training set; (iv) Weights indicating the result
are found via repeated trainings. While it is possible to use all
ANN models as classifiers, the most widely-used model is
MLP with back-propagation NN in medical field. MLP with
back-propagation NNs show a very broad range of applica-
tions like diagnosis of different diseases [22, 26–28], analysis
of EEG [23, 24] and ECG patterns [25, 29] and classification
of Doppler signals [8, 30] in medical field.
Although MLP NN is successfully used in a broad
scope, it is not always the best model among the ANN
models. A MLP NN features the following disadvantages.
(i) When the same data set is used for training and test,
MLP NN produces different outputs in each run because
the MLP NN is run with random initialization of the
weights. This makes difficult reproducibility and compara-
J Med Syst
bility. (ii) Many parameters such as the number of hidden
layers for training of the MLP NN, number of neurons, type
of the activation functions, a set of parameters for learning
function and a probable pruning step must be selected by
the user so that the optimum result can be taken. These
parameters also make the MLP NN algorithm complicated
and make it difficult to coding. Existence of many
additional parameters will extend the training phase. The
adverse check of MLP NN exhibits further difficulty for
requiring the classification margin notion of a target
presented recently [38].
LVQ NN is a nearest-neighbor pattern classifier based on
competitive learning [32]. A LVQ NN has a competitive
(Kohonen) layer and a linear output layer. The linear layer
transforms the classes of competitive layer into user-defined
classifications [31]. LVQ NN structures have been success-
fully used in the tasks of diagnostics of various diseases
[33–35], facial recognition [36] and text classification [37].
LVQ NN applications have superiorities in terms of
more qualitative specifications [33]: (I) LVQ networks have
feature reproducibility because the initialization is more
deterministic. So they produce the same outputs in different
runs, (II) LVQ networks are fast during training; they just
need a few adjusted parameters for training, and can be
retrained with the templates of the local individuals. Also,
since the algorithms are quite simple, application of the
LVQ NN into the software is simple as such (III) Outlier
detection of LVQ NN, which is based on classifying by
distance, can easily be configured to reject patterns that are
not within a certain distance of the prototype patterns, (IV)
LVQ networks can be trained with the unstable dimensions
of various classes. In this study, LVQ NN was used as
classifier for such specifications.
This study basically consists of feature extraction and
classification stages. At the feature extraction stage, power
spectral density (PSD) estimates of internal carotid artery
Doppler signals were obtained by using Burg AR spectrum
analysis technique. At classification stage, LVQ NN was
used to classify internal carotid artery stenosis and healty
subjects by using obtained features with Burg AR spectrum
analysis technique. The classification success which is
observed within the proposed LVQ NN method is compared
with MLP NN, SVM, KNN, NB and C4.5 decision tree
methods’ classification success, based on the sensitivity/
specificity and receiver operating characteristic (ROC) curve
techniques. The classification results showed that the LVQ
NN method was effective for classification of internal carotid
artery Doppler signals.
Rest of the paper is organized as follows. In “Materials
and methods”, by the obtainment of internal carotid artery
Doppler signals, background theory of Burg AR and
LVQ NN methods are explained. The efficiency of the
LVQ NN method, intended for classification of carotid
J Med Syst
artery Doppler signals is demonstrated in “Experimental
results”. “Conclusions”, the paper is concluded with the
discussion section.
Materials and methods
The parts of suggested biomedical system structure are
explained in the following subsections:
Raw data obtaining
Doppler signals were recorded from the 191 subjects, 136
patients as well as of 55 healthy people in Radiology clinics
of Medical Center of Fırat University. Internal carotid artery
examinations were performed with a Doppler unit using a
6.2–8.4 MHz linear transducer. Various tools consist of
Doppler unit (Toshiba SSA-770 Aplio 80, Toshiba, Tokyo,
Japan), an analog/digital interface board and Personal
Computer was used for raw data obtaining. The analog
Doppler unit works in not only continuous mode but also
pulse which wave mode. The probe was most often placed
at an angle of 60° towards the internal carotid artery.
Power spectral density using Burg AR method
Different spectrum estimation methods fall under either
the nonparametric or parametric method categories. In
Nonparametric methods, it is the signal that is used
directly in the estimation of PSD. Periodogram is the
easiest method that can be given as an example to that.
On the other hand, in Parametric methods, a model is
used in the process of estimating the power spectrum.
Burg, Welch, and Yule Walker methods are the most
common parametric methods used.
There are two steps for estimating the spectrum in the
parametric methods. Method parameters are estimated
according to the data series x(n), 0 ≤ n ≤ N−1. PSD
estimate is computed using the estimations obtained here.
As stated above, AR method is preferred as a spectrum
analysis technique, due to the simplification of the
estimation of AR parameters (such as Yule-Walker, Burg,
covariance, least squares, and maximum likelihood esti-
mation). Unlike other AR estimation methods, Burg
method doesn’t calculate the autocorrelation function but
directly estimates the reflection coefficients. Because of
the fact that PSD estimates are close to the true values in
the AR case, to resolve closely spaced sinusoids in signals
with low noise levels and to estimate short data records
can be stated as the main advantages of the Burg’s method
[42]. Moreover, Burg AR method is efficient giving stable
estimates. On the grounds of these advantages, in this
study Burg method is used.
In Burg method, PSD estimation can be defined as
Eq. 1.
^ ðf Þ ¼ e^p


2 ð1Þ

1 þ Pp
P xx
k¼1 a^p ðkÞe
j2pf k

where a^p ðkÞ is AR parameters to reflection coefficients, e^p
is total least-squares error, p denotes model order, f denotes
frequency.
Selecting the model order makes up a key component in
parametric methods. Through various techniques, the
optimal model order can be determined [2, 43, 44]. Akaike
proposed a better criterion for choosing the model order,
namely the Akaike information criterion (AIC) [44]. In
AIC, the model order is selected through the minimization
of Eq. 2.
AICð pÞ ¼ ln s^ 2 þ 2p=N ð2Þ
where s^ 2 is the estimated variance of the linear prediction
error. In this study, model order of the AR method was
taken as 10 by using Eq. 2.
Learning vector quantization
LVQ NN is the nearest-neighbor classifier based on
competitive learning [32]. A LVQ NN has two layers, being
the first competitive and the second linear. The neurons in the
competitive layer are designated as subclasses. Each subclass
has a weight vector. When an input is applied to a LVQ NN,
the best matching is searched for in the competitive layer. The
best matching is designated as the winning neuron. If a
particular neuron in the competitive layer wins, the particular
output of the class of the neuron is set high [36]. The
competitive layer learns the classification of input vectors.
The linear layer transforms the classes of competitive layer
into user-defined classifications. Both competitive and linear
layers have one neuron per class [31].
During the training process of the LVQ NN, the
Euclidean distance from a training vector, x, to each node’s
weight vector, wi, in the competitive layer is computed
according to the Eq. 3.
( )1=2
XN  2
di ¼ kwi
xk ¼ wij
xj ð3Þ
j¼1
The nearest node is declared as the winner, and the
weight vector is adjusted according to whether the winner
node is included in the class of the training vector:
wiþ1 ¼ wi þ aðx
wi Þ; if the winner is the correct class
wiþ1 ¼ wi þ g ðx
wi Þ; otherwise
ð4Þ

Fig. 1 PSD estimations of internal carotid artery Doppler signals for
unhealthy subject suffering from internal carotid artery stenosis
where wi+1 is the weight vector after adjustment, wi the
vector before adjustment, α and γ are learning parameters.
The orientation of the weight adaptation while using
Eq. 4 depends upon whether the class of training pattern
and the class allocated to the reference vector are same. If
they are same, the reference vector is moved near to the
reference vector training template, otherwise, it is moved
away. This movement of the reference vector is controlled
by the learning parameters [39].
Experimental results
In this study, a biomedical system based on LVQ NN has
been developed in order to classify the internal carotid
artery Doppler signals. This study basically consists of
feature extraction and classification stages. At the feature
extraction stage, PSD estimates of internal carotid artery
Doppler signals were obtained by using Burg AR spectrum
analysis technique. LVQ NN was used to classify internal
carotid artery stenosis and healty subjects by using obtained
features with Burg AR spectrum analysis technique. 5 fold
cross validation procedure is preferred in all applications.
Feature extraction and classification processes are imple-
mented by Matlab software package.
Power spectral density estimation of internal carotid artery
Doppler signal
Doppler alteration signals are processed to obtain a Doppler
PSD for the purpose of deriving clinically useful informa-
tion. The Burg AR method offers a qualified spectrum
output for its frequency resolution. Thanks to such better
J Med Syst
frequency resolution, it gets too easy to explain and identify
the activities in the signal using the AR method. In this
study, before the Burg AR spectrum analysis method is
applied to the Doppler signals, sampled Doppler signals are
grouped by frames comprising specific sample numbers.
The most common frame lengths are 64,128 and 256. The
optimal length of the used frame depends upon the
durability of the signal and sampling frequency. In this
study, frame length is taken as 64 because sample numbers
were few [41]. Subsequently, the power spectrum density of
each window was calculated using the Burg AR method.
Hence, number of samples for each subject was reduced to
33. 33 characteristics of the internal carotid artery signals
obtained by calculating the PDSs of Doppler signals were
taken as LVQ NN input parameters.
Figures 1 and 2 present the sample PSDs of the internal
carotid arterial Doppler signals for the healthy subject and the
unhealthy subject respectively. According to Figs.1 and 2,
internal carotid artery Doppler PSD values of healthy
subjects and unhealthy subjects, who suffer from internal
carotid artery stenosis, show different characteristics. There-
fore, such a classification system, being established by taking
such variances in PSD graphics into consideration, enables
for deciding on respective disease.
A statistical package (SPSS) was used in the calculation
of the correlation coefficients for the internal carotid arterial
Doppler signals’ PSD values belonging to the healthy and
unhealthy subjects. Healthy and unhealthy subjects’ corre-
lation coefficient values vary in the range [0.911, 0.976]
and [0.912 0.984] respectively. Positive linear correlations
between the PSD values within each group can be
observed. Based on this result, PSD values of the Doppler
signals are applied to the purposed system.
Fig. 2 PSD estimations of internal carotid artery Doppler signals for
healthy subject
J Med Syst

Classification using LVQ NN

In this study, 5-fold cross validation technique was used.

Therefore, first the training and test sets required for
each fold were determined. The training and test
processes of LVQ NN were conducted using the training
and test sets determined separately for each fold. The
training parameters and the structure of the LVQ NN in
this study were shown in Table 1. The primary factors,
which controlled the behavior of the LVQ NN, are the
number of hidden units and the learning rate. The number
of hidden units has a more serious effect on the stabilization
of the network. It is started up with various hidden
neurons and many tests where each time the percent of
classifier is calculated successfully are performed. For
this application, the best LVQ NN performance was
reached by 20 hidden units trained at the learning rate
of 0.01.
In Fig. 3, training performance of LVQ NN is shown. As
a result of training of the LVQ NNs with the training sets
determined for each fold, it was found out that, mean-
squared error was reduced lower to 0.00654 in an average Fig. 3 The LVQ NN training performance
of 90 epochs. At the end of training periods for 5 fold, LVQ
NN’s achieved a correct classification rate of 100% by
classifying all of the training set correctly. Comparing with other classification methods
After the training stage, the feature vectors belonging to
the test sets for each fold were applied as input to the The classification performance of proposed LVQ NN
trained LVQ NN classifiers. A basic methodology called a method was compared with MLP NN, SVM, C4.5 decision
confusion matrix is used to display the classification results tree, KNN and NB classifier methods. In classification stage,
of a classifier. Labeling the requested classification on the we made experiments by changing the classifiers parameters
rows and the actual classifier outputs on the columns forms in order to find the best classification performances.
a confusion matrix [40]. The confusion matrix showing the The stage of classification with MLP NN, number of
classification results of the LVQ NN is given in Table 2. hidden layers, learning rate, momentum constant value, and
Confusion matrix was created taking into account the test activation functions type modified, with the intent of
errors in all folds. obtaining the best classification performance. Learning rate
According to the confusion matrix in Table 2, 2 was 0.001, momentum constant was 0.9, activation function
subjects with stenosis were incorrectly classified as healthy, type log-sigmoid, mean-squared error 10−5 and maximum
and 2 healthy subjects were incorrectly classified as patients training iteration was determined as 500.
with stenosis. At the classification phase with SVM, radial basis function
was determined as kernel function. The most important
parameters affecting SVM sorting performance are σ param-
Table 1 LVQ NN architecture and training parameters eters those which have C and radial basis function (RBF)
LVQ NN architecture kernel parameters. Parameters of C and σ were determined
The number of layers : 3 on trial way to achieve the best classification performance. C
The number of neuron on the layers: Input : 33 is a regularization parameter between complexity and
Hidden : 20
Output : 2
Activation Functions : Competitive Table 2 Confusion Matrix for the LVQ NN classification results
LVQ NN training parameters
Output/Desired Healthy Stenosis
Learning rate : 0.01
Mean-squared error : 0.00654 Healthy 53 2
Max. Epoch (Stopping criterion) : 1000 Stenosis 2 134
 J Med Syst

Table 3 The statistical parameters of the classifiers methods mance of the classifier using sensitivity and specificity
Classifier Specificity (%) Sensitivity (%) Accuracy (%) values. The ROC curve which is seen Fig. 4 represents the
LVQ NN performance on the test data set. In addition to
LVQ NN 96.36 98.53 97.91 Fig. 4, the areas under the ROC curves were calculated for
SVM 96.23 97.10 96.86 all classifiers in this study. The areas under the ROC curves
KNN 98.15 98.54 98.43 were found to be 0.974 for LVQ NN; 0.956 for SVM, and
Naive Bayes 83.08 99.21 93.75 MLP NN; 0.967 for C4.5; 0.978 for KNN; and 0.950 Naive
C4.5 92.98 98.51 96.86 Bayes classifiers. According to these results, proposed
MLP NN 96.23 97.10 96.86 system can be used confidently to help experts for decision
making in their diagnosis problems.

classification accuracy. The best classification result was Conlusions

obtained for C=16. For the implementation of the SVMs
with the RBF kernel functions, one has to assume a value for In this study, a biomedical system based on LVQ NN has
σ. The optimal σ value was found 0.4. been developed in order to classify the internal carotid
At the phase with classification by means of KNN, the artery Doppler signals. Developed system is mainly
most important parameter affecting classification is k comprised of two stages, namely as being feature extrac-
nearest neighbor number. Usually the optimal value of k tion, and classification. In the feature extraction stage, PSD
is empirically determined. In this study k value was estimates of internal carotid artery Doppler signals were
determined so that it would give the least classification obtained by using Burg autoregressive (AR) spectrum
error (k=3 was determined). In addition, at the phase of analysis technique in order to obtain medical information.
finding of k nearest neighborhood, Euclidean distance was In the classification stage, LVQ NN was used classify
used as distance metric. features from Burg AR method. In experiments, LVQ NN
C4.5 decision tree algorithms consist of two distinct based method reached 97.91% classification accuracy with
phases: building phase and pruning phase. In the building 5 fold CV technique. The classification results of LVQ NN
phase, initially, the decision tree had a single root node for method were compared with the classification performance
the entire training set. In the pruning phase post-pruning of MLP NN, SVM, KNN, decision tree and NB methods
method was used to decide when to stop expanding a using sensitivity/specificity values and ROC analysis.
decision tree. The confidence factor is used for pruning the According to the comparison results, LVQ NN is a
tree. In this study, confidence factor assigned is 0.25. The reasonable classifier system, and effective for classification
pruned tree consisted of 4 leaves and 7 nodes.
The NB classifier is a classification method that is based on
probability calculation. It assumes that all the attribute values
are conditionally independent simplifying its calculation. A
new case is classified into the class having the highest
calculated probability. In NB classification, the Laplace-
estimate was used for estimation of prior probabilities.
The classification performance of each method was
evaluated using the statistical parameters like sensitivity and
specificity.
Sensitivity: number of true positive decisions/number
of actually positive cases.
Specificity: number of true negative decisions/number
of actually negative case.
Table 3 presents comparison results depending on
sensitivity/specificity rates. Experimental results show the
performances of these different classifiers are close and
comparable.
A ROC graph is a technique for visualizing, organizing
and selecting classifiers based on their performance. The
ROC curve was plotted in order to compare the perfor- Fig. 4 ROC curve for LVQ NN classifier
J Med Syst
of internal carotid artery Doppler signals. Overall LVQ NN
can be used confidently to help experts for decision making
in their diagnosis problems.
Acknowledgements The authors acknowledge the support of this
study provided by Selçuk University Scientific Research Projects.
References
1. Stoica, P., and Moses, R., Introduction to spectral analysis.
Prentice-Hall, New Jersey, 1997.
2. Kay, S. M., Modern spectral estimation: theory and application.
Prentice-Hall, New Jersey, 1988.
3. Baker, W. H., Diagnosis and treatment of carotid artery disease.
Futura Publishing Company Inc., New York, 1985.
4. Wright, I. A., Gough, N. A. J., Rakebrandt, F., Wahab, M., and
Woodcock, J. P., Neural network analysis of Doppler ultrasound
blood flow signals: A pilot study. Ultrasound Med. Biol. 23
(5):683–690, 1997.
5. Evans, D. H., McDicken, W. N., Skidmore, R., and Woodcock, J.
P., Doppler ultrasound: Physics. Instrumentation and clinical
applications. Wiley, Chichester, 1989.
6. Übeyli, E. D., and Güler, İ., Comparison of eigenvector methods with
classical and model-based methods in analysis of internal carotid
arterial Doppler signals. Comput. Biol. Med. 33(6):473–493, 2003.
7. Güler, İ., Kıymık, M. K., Akın, M., and Aklan, A., AR Spectral
analysis off EEG signals by using maximum likelihood estima-
tion. Comput. Biol. Med. 31:441–450, 2001.
8. Übeyli, E. D., and Güler, İ., Neural network analysis of internal
carotid arterial Doppler signals: Predictions of stenosis and
occlusion. Expert. Syst. Appl. 25:1–13, 2003.
9. Übeyli, E. D., and Güler, İ., Spectral analysis of internal carotid
arterial Doppler signals using FFT, AR, MA, and ARMA
methods. Comput. Biol. Med. 34(4):293–306, 2004.
10. Übeyli, E. D., and Güler, İ., Selection of optimal AR spectral
estimation method for internal carotid arterial Doppler signals
using Cramer-Rao bound. Comput. Electr. Eng. 30(7):491–508,
2004.
11. Özşen, S., Kara, S., Latifoğlu, F., and Güneş, S., A new
supervised classification algorithm in artificial immune systems
with its application to carotid artery Doppler signals to diagnose
atherosclerosis. Comp. Methods Programs Biomed. 88:246–255,
2007.
12. Kara, S., and Dirgenali, F., A system to diagnose atherosclerosis
via wavelet transforms, principal component analysis and artificial
neural networks. Expert Syst. Appl. 32(2):632–640, 2007.
13. Übeyli, E. D., Statistics over features for internal carotid arterial
disorders detection. Comput. Biol. Med. 38:361–371, 2008.
14. Ceylan, M., Ceylan, R., Dirgenali, F., and Özbay, Y., Classifica-
tion of carotid artery Doppler signals in the early phase of
atherosclerosis using complex-valued artificial neural network.
Comput. Biol. Med. 37(1):28–36, 2007.
15. Özbay, Y., and Ceylan, M., Effects of window types on
classification of carotid artery Doppler signals in the early phase
of atherosclerosis using complex-valued artificial neural network.
Comput. Biol. Med. 37(3):287–295, 2007.
16. Polat, K., Kara, S., Latifoğlu, F., and Güneş, S., Pattern detection
of atherosclerosis from carotid artery Doppler signals using fuzzy
weighted pre-processing and least square support vector machine
(LSSVM). Ann. Biomed. Eng. 35(5):724–732, 2007.
17. Güler, İ., and Übeyli, E. D., A recurrent neural network classifier
for Doppler ultrasound blood flow signals. Pattern Recogn. Lett.
27(1):1560–1571, 2006.
18. Güler, İ., and Übeyli, E. D., Implementing wavelet/probabilistic
neural networks for Doppler ultrasound blood flow signals. Expert
Syst. Appl. 33(1):162–170, 2007.
19. Dirgenali, F., and Kara, S., Recognition of early phase of
atherosclerosis using principles component analysis and artificial
neural networks from carotid artery Doppler signals. Expert Syst.
Appl. 31(3):643–651, 2006.
20. Kemaloğlu, S., Erdoğan, N., and Kara, S., Discontinuous doppler
signals simulating respiratory misregistration: Effect on autoregres-
sive frequency spectra. Comput. Biol. Med. 36(5):465–472, 2006.
21. Ölmez, T., and Dokur, Z., Classification of heart sound using an
artificial neural network. Pattern Recogn. Lett. 24:617–629, 2003.
22. Türe, M., and Kurt, İ., Comparison of four different time series
methods to forecast hepatitis A virus infection. Expert Syst. Appl.
31(1):41–46, 2006.
23. Walczak, S., and Nowack, W. J., An artificial neural network
approach to diagnosing epilepsy using lateralized bursts of theta
EEGs. J. Med. Syst. 25(1):9–20, 2001.
24. Subaşı, A., Application of classical and model-based spectral
methods to describe the state of alertness in EEG. J. Med. Syst. 29
(5):473–486, 2005.
25. Sinha, R. K., Aggarwal, Y., and Das, B. N., Backpropagation
artificial neural network detects changes in electro-encephalogram
power spectra of syncopic patients. J. Med. Syst. 31:63–68, 2007.
26. Hayashi, Y., and Setiono, Y., Combining neural network pre-
dictions for medical diagnosis. Comput. Biol. Med. 32(4):237–
246, 2002.
27. Er, O., and Temurtaş, F., A study on chronic obstructive
pulmonary disease diagnosis using multilayer neural networks.
J. Med. Syst. 32:429–432, 2008.
28. Çınar, M., Engin, M., Engin, E. Z., and Ateşçi, Y. Z., Early
prostate cancer diagnosis by using artificial neural networks and
support vector machines. Expert Syst. Appl. 36(3):6357–6361,
2009.
29. Moavenian, M., and Khorrami, H., A qualitative comparison of
Artificial Neural Networks and Support Vector Machines in ECG
arrhythmias classification. Expert Syst. Appl. 37(4):3088–3093,
2010.
30. Yıldırım, H., Altınsoy, H. B., Barışçı, N., Ergün, U., Oğur, E.,
Hardalaç, F., and Güler, İ., Classification of the frequency of
carotid artery stenosis with MLP and RBF neural networks in
patients with coroner artery disease. J. Med. Syst. 28(6):591–601,
2004.
31. Demuth, H., and Beale, M., Neural netwoks toolbox for use with
MATLAB. The MathWorks Inc., Massachusetts, USA, 1994.
32. Bashar, M. K., Ohnishi, N., Matsumoto, T., Takeuchi, Y., Kudo,
H., and Agusa, K., Image retrieval by pattern categorization using
wavelet domain perceptual features with LVQ neural network.
Pattern Recogn. Lett. 26:2315–2335, 2005.
33. Dieterle, F., Müller-Hagedorn, S., Liebich, H. M., and Gauglitz,
G., Urinary nucleosides as potential tumor markers evaluated by
learning vector quantization. Artif. Intell. Med. 28:265–279, 2003.
34. Temurtaş, F., A comparative study on thyroid disease diagnosis
using neural networks. Expert Syst. Appl. 36:944–949, 2009.
35. Kara, S., and Güven, A., Training a learning vector quantization
network using the pattern electroretinography signals. Comput.
Biol. Med. 37:77–82, 2007.
36. Bashyal, S., and Venayagamoorthy, G. K., Recognition of facial
expressions using Gabor wavelets and learning vector quantiza-
tion. Eng. Appl. Artif. Intell. 21:1056–1064, 2008.
37. Martin-Valdivia, M. T., Urena-Lopez, L. A., and Garcia-Vega, M.,
The learning vector quantization algorithm applied to automatic
text classification tasks. Neural Netw. 20(6):748–756, 2007.
38. Allwein, E. L., Schapire, R. E., and Singer, Y., Reducing
multiclass to binary: a unifying approach for magrin classifiers.
J. Machine Learn Res. 1:113–41, 2000.

39. Masouma, S., Bouveresse, D. J. R., Vercauteren, J., Jalali-Heravi,
M., and Rutledgeb, D. N., Discrimination of wines based on 2D
NMR spectra using learning vector quantization neural networks
and partial least squares discriminant analysis. Anal. Chim. Acta
558:144–149, 2006.
40. Beksaç, M. S., Başaran, F., Eskiizmirliler, S., Erkmen, A. M.,
and Yörükan, S., A computerized diagnostic system for the
interpretation of umbilical artery blood flow velocity wave-
forms. Eur. J. Obstet. Gynecol. Reprod. Biol. 64(1):37–42,
1996.
J Med Syst
41. Akın, M., and Kıymık, M. K., Application of periodogram and AR
spectral analysis to EEG signals. J. Med. Syst. 24(4):247–256, 2000.
42. Faust, O., Acharya, R. U., Allen, A. R., and Lin, C. M., Analysis
of EEG signals during epileptic and alcoholic states using AR
modeling techniques. IRBM 29(1):44–52, 2008.
43. Proakis, J. G., and Manolakis, D. G., Digital signal processing
principles, algorithms, and applications. Prentice-Hall, New
Jersey, 1996.
44. Akaike, H., A new look at the statistical model identification.
IEEE Trans. Autom. Control 19:716–723, 1974.