313
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
OPTIMAL CHOICE OF SUPERVISED TECHNIQUES FOR
MR IMAGE CLASSIFICATION
Balasubramanian Aruna Devi
Electronics and Communications engg,
Kalasalingam Academy of research and Education,
Krishnankoil, (India).
E-mail: b.arunadevi@klu.ac.in ORCID: https://orcid.org/0000-0002-0981-804X
Murugan Pallikonda Rajasekaran
Professor. Electronics and Communications engg,
Kalasalingam Academy of research and Education
Krishnankoil, (India).
E-mail: m.p.raja@klu.ac.in ORCID: https://orcid.org/0000-0001-6942-4512
Recepción:
05/12/2019
Aceptación:
23/12/2019
Publicación:
23/03/2020
Citación sugerida:
Devi, B. A., y Rajasekaran, M. P. (2020). Optimal choice of supervised techniques for MR image
classication. 3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020, 313-327.
http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
Suggested citation:
Devi, B. A., & Rajasekaran, M. P. (2020). Optimal choice of supervised techniques for MR image
classication. 3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020, 313-327.
http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
314 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
ABSTRACT
Magnetic Resonance Imaging (MRI) is a modern, robust method that uses in the detection
of various medical problems. In this research work, a trial is used to attempt for the detection
of tumour in pancreas MR images. An automated classier is used for detection of tumour
in MR images and avoids the drawbacks of MRI. This automated classiers can detect
automatically, either the MR image is aected or not aected. Features are extracted from
MR images using second order statistics approach and are classied by two techniques
Support Vector Machine (SVM) and Extreme Learning Machine (ELM). SVM approach
has high classication accuracy (96%) which is higher than ELM, while ELM performs
faster compared to SVM.
KEYWORDS
SVM, ELM, GLCM feature extraction, Image classication.
315 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
1. INTRODUCTION
In medicine eld, medical Image analysis and processing has enormous applications. It
has emerged as one of the superior tools to diagnose as well as detect many disorders. It
permits both radiologists and doctors to make exact detection, by analyzing and visualizing
the medical image. Computer-Aided Diagnosis (CAD) is an approach that is achieving
attention in modern life. It can comfort doctors accurately read images and diagnosis
possible decisions to avoid incorrect understanding of lesions. It is necessary to mark that
CAD systems can only present a second opinion and can by no means follow physicians
or radiologists. There are many imaging modalities for the humans of tissue analysis, such
as Magnetic Resonance Imaging (MRI), mammogram function, Computed Tomography
(CT) and so on. The main target of this research work is on MRI images. MRI (Armstrong,
Cohen, Weinbrg, & Gilbert, 2004) is a medical imaging method that generates images of
the inner part of human body. It is an on-radio active, non-aggressive, pain-free method for
visualizing detailed data regarding the normal or tumors without any human involvement.
The target of this research work is to grant an automatic detection tool that will guide
physicians or radiologists in detecting lesions by diagnosing them from normal tissue. The
rst step is to extract the features in MRI Images by second order statistics. These extract
important image features from the MR Images are used to classify the image is aected or
not aected. This will help the physicians or radiologists in the analysis of diagnosing tumor
in MR images. In this research work, we have analyzed two classiers such as SVM (Vapnik,
1995) & ELM (Huang, Zhu, & Siew, 2004).
SVM
SVM is a classication approach for high-dimensional data which is presented by Vapnik
(1995) to resolve the discrimination disputes of two issues. SVM has been broadly used in
the elds of medical image processing, image retrieval, text analysis, and so on. SVM is
based on the working principle that the data in the input space can be linear dividable in a
higher dimensional feature space after a certain mapping.
316 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
ELM
ELM (Huang et al., 2004) is a newly advanced machine learning approach, extensively
applied in image processing, computer Vision, biomedical applications, system modeling
and regression.
The rst stage describes with the feature extraction applying GLCM in MRI pancreas
images. In the second stage, they are discriminated by the classication methods such as
SVM & ELM. The output displayed that SVM classication method has the superior
classication accuracy compared to ELM.
2. LITERATURE SURVEY
Lu et al. (2017) presented DWT feature extraction and classied by bat algorithm based
extreme learning machine with a classication accuracy of 93.33%. Nazir, Wahid, and
Khan (2015) classied brain MRI using various moment features extraction and articial
neural networks with an accuracy of 94.24%. Nandpuru, Salankar, and Bora (2014)
classied brain tumor applying Texture features, symmetrical and gray features extraction,
principal component analysis (PCA) feature selection and support vector machine (SVM)
classication. The classication accuracy was 84%. Ibrahim, Osman, and Mohamed (2018)
classied MR brain images using wavelet-based features extraction, features selection by
PCA, and classied by articial neural networks (ANN) with a classication accuracy of
96.33%. Othman, Abdullah, and Kamal (2011) discriminated normal and abnormal MRIs
using DWT feature extraction, principal component analysis (PCA) feature selection and
SVM classication by 65% classication accuracy. Kavitha and Thyagharajan (2012) have
presented histogram, textural features and classied by SVM with an accuracy of 90%. Diz,
Marreiros and Freitas (2015) have described GLCM and Grey-Level Run Length Matrix
(GLRLM) feature extraction for mammogram image classication and achieved 76%
accuracy. Dheeba and Selvi (2011) have presented Laws texture features to discriminate
images into Benign and Malignant (MIAS-Mammographic Image Analysis Society
database) and gained 86.10% accuracy. Shah, Surve and Turkar (2015) classied pancreatic
tumor of CT images using Minimum distance classier. The classication accuracy was
61.59%. Yao, Chen and Chow (2009) described wavelet transform features extraction
method and classied by SVM with an accuracy of 83%. Aruna Devi and Pallikonda
317 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Rajasekaran (2018) described GLCM features extraction and dierentiated normal and
abnormal MR images by applying ANN and SVM methods. ANN found that it achieves
96% classication accuracy. Aruna Devi, Pallikonda Rajasekaran and Thiyagarajan (2019)
proposed tumor discrimination by GLCM feature extraction, JAFER feature selection
and comparison among 5 types of classication modes, ANN BP gains 98% classication
accuracy. Based on the survey, the discrimination capability of classiers is very less and
computational time is also high. Our proposed technique increases the discrimination
capability as high (98%) that is superior to previous research analysis. In this research work,
we are going to discriminate the 160 MR images by aected or not aected using second
order statistics feature extraction and discriminated by two modes such as ANN and SVM.
Sensitivity, specicity and classication accuracy is measured and compared among the two
modes.
3. MATERIALS AND METHODS
Input data set:
The dataset used for predicting the performance of the proposed model in this research
work is based on the MR pancreas medical images that are gathered from the health care
centres. The numbers of medical pancreas images totally 160 of which 100 are normal and
60 are abnormal images. Figure 1 shows the normal pancreas images and Figure 2 shows
the abnormal pancreas images. Figure 3 displays the owchart.
4. FEATURES EXTRACTION
Features extraction by second order statistics:
The procedure of transferring the input image into a set of features is known as feature
extraction. Features normally consist of data relevant to colour, shape, texture or context.
First order statistics provides gray level pixels occur in an image. First order measures are
mean, variance, skewness and kurtosis. Second order statistics provides inter relationship
between pixel and its neighbors. They provide detailed information about the pixel and
its neighbors with an angle of 0, 45, 90 and 135 degrees at a distance d. Second order
measures are entropy, energy, contrast, homogeneity, sum of variance, cluster prominence,
318 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
sum of entropy, cluster shade, information measure of correlation. Second order statistics
features are tabulated in Table 1. In this work, these features are measured and classied by
two classiers namely SVM and ELM.
a b c
d e f
g h i
Figure 1. Normal pancreas MR images.
a b c
d d f
319 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
g h i
Figure 2. Abnormal pancreas MR images.
Features extraction using second order statistics
Classifiers
Decision
SVM ELM
Affected Not affected
Figure 3. Flow chart.
Table 1. GLCM features.
GLCM features:
Contrast:
(1)
Energy: (2)
Entropy:
(3)
Homogeneity: (4)
320 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
GLCM features:
Cluster prominence: It denes the inequality of an image.
Cluster shade: It is exact to cluster prominence in that it also denes the loss of similarity in the image.
Sumentropy:
(5)
Sumvariance:
(6)
Information measure of correlation: HXY-HXY1/max{HX,HY};
(7)
5. CLASSIFICATION
Classication is the procedure of classifying a known input by a good classier. The main
target of proper classication is to provide a label to each MR image based on second order
statistics features.
SVM
SVM is used for mapping complex feature space into linear feature space. It works on the
base of tting a boundary to a eld of points that are belongs to one class with one another.
Once boundary is xed, on the learned samples, for any unknown points that are test
sample need to be classied, and the accuracy will be predicted. Once boundary is xed,
maximum training points are redundant. All it demands a group of points that can identify
and ts the boundary. The group of points are known as support vectors and the boundary
is called as hyperplane.
ELM
Extreme Learning Machine (ELM) is a single hidden-layer feed-forward neural network
(SLFN). The signicance of the SLFN should be convenient for information such as
weight, threshold value, and activation function so that superior training can be achieved.
In gradient-based learning, all of these quality measures are changed iteratively for the
signicant value. Therefore, due to the possibility of being attached to the slow and local
minimum, the performance can generate low outputs. On the basis of the gradient in
the ELM training process, the output weights are analytically calculated where the input
321 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
weights are chosen randomly. In this process, the success rate raises because the resolution
time and the errors can actively shorten the possibility of being tted to a local minimum.
6. TRAINING AND TESTING
Feature dataset is classied by the classication such as SVM & ELM techniques. Out of
165 pancreas MR images, 70% Images were used for learning and 30% of images were
used for testing.
There are more classication techniques, where training dataset are favoured by random
sampling with restoration for MR pancreas image classication. Basically, choosing the
best feature election approach with its performance parameters, it was required for process
of classication approaches. In continuity, the classication outcomes with parameters for
each discrimination approach are followed. Performance parameters were employed for
each discrimination approach and the best one was choose for tumour identication.
Support vector machine is relevant for high dimension small-sample learning and nonlinear
problems. SVM mainly focus at binary classication. SVM provides strong generalization
ability and structural risk minimization. It separates the two classes using hyper plane. The
second order statistics features are used as input of the classier and the corresponding
known label (that is aected or not aected) is the output of the classier. The hyper plane
separates the class as aected or not aected. It helps to train the classier as input, output
mappings functions. It provides the superior minimum distance to the learning data. The
SVM has many kernel functions. The predominant kernel function is RBF that denotes
radial basis function. Here RBF kernel is used. After learning the classier, the test set was
applied to test the signicance of the classier and its capability to accurately discriminate
the MR images as either aected or not aected. To check out our SVM classier, a
confusion matrix was generated as shown in Figure 4 and the classication accuracy is 96%
which is tabulated in Table 2.
ELM
Conventional single hidden-layer feed forward neural networks (SLFNs), such as the back
propagation (BP) method, have been applied for research in many applications. The weight
which assigns the hidden nodes is applied randomly and weights are not altered forever.
322 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
The weights that correlate hidden nodes with outputs are trained in one step. ELM yields
cooperative training technique. Even though ELM conrms to be terribly quick and in a
great way in observation, it has some drawbacks. The election of hidden neurons is placed
on trial and error technique that can provide into inadequate results. ELM provides 89%
classication accuracy.
Next, the performance evaluation approaches applied to test the two methods (SVM &
ELM). We analyse the execution of the methods in terms of sensitivity, specicity and
accuracy.
Performance opinion parameters:
Sensitivity = TP/ (TP +FN) 100%
S p e c i c i t y = T N / ( T N + F P ) 1 0 0 %
Accuracy = (TP +TN)/ (TP+TN+FP+FN) 100%
Where:
TN (True Negative) = perfectly discriminated negative cases, TP (True Positives) = perfectly
discriminated positive cases, FN (False Negative) = imperfectly discriminated positive cases
FP (False Positives) = imperfectly discriminated negative cases.
Specicity measures how perform the system can predict the negatives. Sensitivity is the
rate of perfectly discriminated positives, describes best performance of the approach in
predicting positives. Accuracy conrms the whole correctness of the classier in predicting
both positive and negative cases in terms of tumor.
7. RESULTS
SVM and ELM are learned by second order statistics features and classify the pancreas
MR image as aected or not aected (normal or abnormal). SVM method provides good
accuracy than ELM. The accuracy of SVM method is 96%, specicity and sensitivity
are 95 %and 97% respectively. The classication accuracy of ELM is 89%, specicity
and sensitivity are 95%and 92% respectively. Table 2 represents classication accuracy,
sensitivity and specicity for two methods. Figure 4 represents SVM confusion matrix.
Figure 5 represents ELM ROC curve is graphed as a plot of true-positive rate on the y-axis
and false-positive rate on the x-axis.
323 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
8. CONCLUSION
Among the two methods, SVM is the superior method which gives 96% classication
accuracy. It proves that our proposed SVM technique gives high accuracy of 96%, which
is compared to other previous techniques. In this work, using second order stastics features
are extracted and classied by SVM and ELM. Results are compared and proved SVM
provides high classication accuracy of 96%. In future, best features are selected using
feature reduction methods (forward selection, backward elimination) and are classied by
SVM and other classiers to realize which classier is best in practice.
9. FUTURE WORK
In this work, using second order statistics features are extracted and classied by SVM
and ELM. Results are compared and proved SVM provides high classication accuracy
of 96%. In future, best features are selected using feature reduction methods (forward
selection, backward elimination) and are classied by SVM and other classiers to realize
which classier is best in practice.
Table 2. Comparative analysis of SVM and ELM techniques.
Classication
Techniques
Classication
accuracy
Sensitivity Specicity Time
SVM 96.67 97.3 95.65 1.020s
ELM 89.05 92.3 95.24 0.320s
Figure 4. SVM confusion matrix.
324 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Figure
5.
ELM ROC.
ACKNOWLEDGEMENT
The authors thank the management of Kalasalingam Academy of research and education
for granting nancial assistance underside the rule of University Research Fellowship
(URF) in Department of Electronics and Communication Engineering. Also, we thank
KGS health care centre, Madurai for granting the pancreas MR image that is very useful
for this research work
325 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
REFERENCES
Armstrong, T. S., Cohen, M. Z., Weinbrg, J., & Gilbert, M. R. (2004). Imaging
techniques in neuro oncology. Seminars in Oncology Nursing, 20(4), 231-239. https://
doi.org/10.1016/j.soncn.2004.07.003
Aruna Devi, B., & Pallikonda Rajasekaran, M. (2018). Performance Evaluation
of MRI Pancreas Image Classication Using Articial Neural Network (ANN).
In Satapathy S., Bhateja V., Das S. (eds.) Smart Intelligent Computing and Applications.
Smart Innovation, Systems and Technologies, vol. 104. Springer, Singapore. https://doi.
org/10.1007/978-981-13-1921-1_65
Aruna Devi, B., Pallikonda Rajasekaran, M., & Thiyagarajan, A. P. (2019). Analysis
and classication of malignancy in pancreatic magnetic resonance images using
neural network techniques. International journal of imaging systems and technology. https://
doi.org/10.1002/ima.22314
Huang, G., Zhu, Q., & Siew, C. (2004). Extreme learning machine: a new learning scheme
of feedforward neural networks. In 2004 IEEE International Joint Conference on Neural
Networks (IEEE Cat. No.04CH37541), 2, 985-990, vol. 2. https://doi.org/10.1109/
IJCNN.2004.1380068
Ibrahim, W. H., Osman, A. A., & Mohamed, Y. I. (2013). MRI brain image
classication using neural networks. In 2013 International Conference on Computing,
Electrical and Electronic Engineering (ICCEEE), 253-258. https://doi.org/10.1109/
ICCEEE.2013.6633943
Kavitha, S., & Thyagharajan, K. K. (2012). Features based mammogram image
classication using weighted feature support vector machine. In Krishna P. V., Babu
M. R., Ariwa E. (eds.) Global Trends in Information Systems and Software Applications.
ObCom 2011. Communications in Computer and Information Science, vol. 270. Springer,
Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-29216-3_35
326 http://doi.org/10.17993/3ctecno.2020.specialissue4.313-327
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Lu, S. S., Qiu, X., Shi, J., Li, N., Lu, Z.H., Chen, P., … Zhang, Y. D. (2017). A
pathological brain detection system based on extreme learning machine optimized
by bat algorithm. CNS & neurological disorders drug targets, 16(1), 23–29. https://doi.org
/10.2174/1871527315666161019153259
Nandpuru, H. B., Salankar, S. S., & Bora, V. R. (2014). MRI brain cancer classication
using Support Vector Machine. In 2014 IEEE Students’ Conference on Electrical, Electronics
and Computer Science, 1-6. https://doi.org/10.1109/SCEECS.2014.6804439
Nazir, M., Wahid, F., & Khan, S. A. (2015). A simple and intelligent approach for brain
MRI classication. Journal of Intelligent and Fuzzy Systems, 28, 1127-1135. https://doi.
org/10.3233/IFS-141396
Othman, M. F., Abdullah, N., & Kamal, N. F. (2011). MRI brain classication using
support vector machine. In 2011 Fourth International Conference on Modeling, Simulation
and Applied Optimization, 1-4. https://doi.org/10.1109/ICMSAO.2011.5775605
Shah, J., Surve, S., & Turkar, V. (2015). Pancreatic Tumor Detection using Image
processing Techniques. Procedia Computer Science, 49, 11-16. https://doi.org/10.1016/j.
procs.2015.04.221
Vapnik, V. (1995). The Nature of Statistical Learning Theory. New York: Springer.
Yao, J., Chen, J., & Chow, C. (2009). Breast Tumor Analysis in Dynamic Contrast Enhanced
MRI Using Texture Features and Wavelet Transform. IEEE Journal of Selected Topics in
Signal Processing, 3, 94-100. https://doi.org/10.1109/JSTSP.2008.2011110
