mri based tumor detection using machine learning

mri based tumor detection using machine learning

1. Detection and Classification of Brain Tumor using
Support Vector Machine Based GUI

Imran Ullah Khan Shamim Akhter Shaheen Khan
Department of Electronics and Department of Electronics and Department of Electronics and
Communication Engineering Communication Engineering Communication Engineering
Integral University Jaypee Institute of Information Mewat Engineering College
Lucknow, India
Technology, NOIDA, India Haryana, India
[email protected]
[email protected] [email protected]
Abstract- Medical image segmentation is a challenging
task in the field of medical science. Many tools have been
developed by engineers to detect tumor and perform
analysis of medical images. The most important and
effective role in the entire procedure is played by image
segmentation tool. It has attracted a lot of attention in the
last so many years and researchers are continuously
working to increase its quality and attributes. This paper is
about the detection of brain tumor using a support vector
machine based interface using GUI in Matlab. The
interface can use any combination of segmentation, filtering
and other techniques to achieve optimum results. The
algorithm begins with noise removal and feature extraction
using discrete wavelet transform. The extracted features
include both first and second order features. These features
are reduced to the desired level using principle component
analysis. These features are also used to train the kernel
SVM. The classification is then performed by support
vector machine. The interface of GUI is developed using
Matlab guide.
Keywords-
Matlab, Support vector machine, GUI.
I. INTRODUCTION
In the present world,image processing methods
involved in the digital biomedical area holds an
important position in two major areas [1]. These areas
include improvement in the pictorial information for the
purpose of human studies and processing of this data for
storage [2]. The analysis of the data available in the form
of images decides the success of the respective task
being performed. The manual analysis is time consuming
and more error prone. Hence the automation of the
analysis process is very significant these days. This
incorporation of automation process in medical science
to develop a tool for diagnosis is certainly a boon for
mankind. Moreover the automated tool is more accurate
and reliable than the human readers. The imbalance
between the growth rate and the death rate of cells
results in the formation of tumor. When this kind of
activity occurs in brain then it is termed as brain tumor.
According to the data gathered by National Brain Tumor
Foundation (NBTF), brain tumor is termed as the most
harmful disease in the past two decades [3].
There are basically two categories of tumor, benign and
malignant brain tumor. The benign brain tumor cells are
non-invasive. They only destroy the affected area and the
remaining body parts do not have any effect of these
cells. Though, these can cause other serious problems.
Once removed, benign tumor rarely grows back. Brain
cancer is another name for malignant brain tumors.
These are invasive in nature and have a very swift
growth [4]. The problem with brain tumor is that there is
nothing specific about its symptoms. It is often mistaken
for any other common disease. This can cause delay in
the diagnosis which increases the risk factor. The
detection at an early stage is very important because then
the required and proper diagnosis can follow. The
symptoms include nausea, vomiting, severe headache,
speech problems, vision impairment, and hearing
impairment, problem in walking and seizures [5]. The
stage at which the tumor is identified plays a vital role in
the follow up diagnosis. The patient undergoes multiple
scans using various technologies to take the picture of
the brain.There are many image modalities that can be
used forthe purpose of scanning the required body part.
These include Magnetic Resonance Imaging (MRI),
Computed Tomography (CT), Single Photon Emission
Computed Tomography (SPECT), Magnetic resonance
Spectroscopy (MRS) and others. Amongst all these MRI
has emerged as the primary choice of the surgeons these
days to get the brain scan of the patients. This high
resolution technique is utilized in the radiology
department to obtain the details about the internal
structures of the body. It is suitable for the brain as it is
very sensitive and non-invasive [6]. The scan is obtained
by increasing the contrast discrimination and can be
acquired in any plane. It also helps in determining the
precise location of the tumor which is very important
from the diagnosis point of view.
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
978-1-7281-5475-6/20/$31.00 ©2020 IEEE 739

2. To identify and classify brain tumor, Computer
Aided Diagnosis (CAD) systems are developed with the
help of MR images. The classification task can be
performed based on the variety of features obtained from
the images. These features include first order features,
second order features, shape and texture features [7]. The
aim of this work is to extract and group together the best
feature sets from the scan with the help of principle
component analysis to achieve optimum results. The first
and second order features are extracted from the MR
images and fed to the classifier. This classifier will
generate a training file for the purpose of evaluating the
test data. The next section describes some of the
literature work related to brain tumor classification.
II. LITERATURE REVIEW
There are many classification methods available in
the literature to detect and identify brain tumor. Support
Vector Machine (SVM) is notably considered a good one
for a two class classification purpose. It provides better
results for noisy data. It is a supervised learning method
which makes the analysis less complicated and is used
for data analysis and pattern recognition. SVM can
perform data analysis by either classification or
regression. This work is based on considering image
segmentation as a classification problem. The lesion in
the brain is classified as benign or malignant using SVM
in this project.
The MR images can be classified using both
supervised as well as unsupervised techniques, for
instance artificial neural networks, self-organization
map, fuzzy c-means, k-nearest neighbours and many
others. The classification can be based on feature sets or
forming similar trait pixels. A large amount of work is
found in the literature related to the same. A few papers
are listed below.
P. Kumar et Al [8] suggested a new method for
automatic dissection of brain neoplasms in MR images.
Their method consists of four stages - preprocessing,
dissection, feature extraction, feature reduction and
classification. They have employed Weiner filter to
eliminate the noise. Features are extracted using
histogram and Co-occurrence matrix and further the
dimensionality of the features are reduced using PCA.
For classification a kernel based SVM was used.
Shankaragowda B.B. et Al [9] exploited the same
technique of preprocessing, segmentation, feature
extraction and classification with some changes here and
there. This method classifies the MR image as benign
and malignant. The method is based on SVM classifier.
The analysis became quicker with this method and it
eliminates the need of biopsies.
KetanMachhale et Al [10] adopted an intellectual
classification approach to classify the brain MR images
as normal and abnormal. Image preprocessing, image
feature extraction and classification were used for the
respective purpose. They have compared different
machine learning techniques: Support Vector Machine
(SVM), K- Nearest Neighbor (KNN) and Hybrid
Classifier (SVM-KNN) is used to classify 50 images.
The highest accuracy was achieved using hybrid
classifier (SVM –KNN).
Nan Zhang et Al [11] used the multi kernel SVM
classification along with the fusion process to detect
brain tumor from multi sequence MR images. This
method consists of two steps. The first step is to classify
the tumor region using a multi-kernel SVM which
performs on multi-image sources and obtains relative
multi-result. The second step ameliorates the contour of
the tumor region using both the distance and the
maximum likelihood measures.
HariBabuNandpuru et Al [12] used classification
techniques based on SVM to classify the brain MR
image as normal or abnormal. They have used gray scale
symmetrical and texture features for feature extraction.
Three kernel functions – linear, quadratic and
polynomial were employed with SVM. The SVM
achieved the highest accuracy with quadratic kernel,
about 84%.
Vijay Wasule et Al [13] used the SVM-KNN classifier to
classify the brain MR image into malignant or benign
and low grade or high grade. GLCM technique was used
for feature extraction and stored as feature vectors.
Entropy, energy, homogeneity, correlation and other
features are calculated. The accuracy was about 96% and
86% for SVM and KNN respectively.
While some scholars tried to improve the
accuracy and precision of the result, some tried to reduce
the time taken during the process. The progress has
increased with time and there is a lot of scope to achieve
much more in this field. In this work the classification
method is used to classify the tumor as benign or
malignant by developing an interface based on support
vector machine. The analysis is done by extracting
different feature sets from the MR scan and optimizing
the features to obtain better results. The features are
displayed in the interface.
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
740

3. The rest of the paper is divided as follows: section 3
describes the proposed method, data sets, features
extracted from the images, the classifier, and the
interface, section 4 describes the experimental work and
section 5 concludes the work and gives future aspects in
the related field.
III. PROPOSED METHOD
The work is initiated by obtaining the MR scans of
the patients for classification and analysis purpose. All
the images are divided into training and testing sets. The
training set is analysed first by extracting first and
second order features. These features are then used by
the classifier to generate the required file. These features
from the test images are then used to identify and
classify the tumor as benign or malignant. An interface is
prepared based on this algorithm to make the whole
procedure easy and fast. The flow chart of the work is
shown in figure 1.

Figure 1: Flowchart of the proposed method
A. DATASET
The MRI scans of patients with benign and
malignant brain tumor are collected from The Cancer
Imaging Archive public access [14]. Some normal scans
are also obtained for validation purposes. This dataset is
used to perform classification using SVM. The size and
location of the lesion are well evident in the images. The
sample images are shown below in figure 2.

Figure 2: Normal and Tumorous brain MRI scans
The images obtained for the classification
purpose are grouped into two different sets for training
and testing the SVM. The dataset consists of a total of
110 images including the normal brain scans. The
division of images is shown in table 1 below.
Table 1: Division of training and test images
Images Training Validation
110 88 22
10(N) 77(AN) 4(N) 18(AN)

B. FEATURE EXTRACTION AND REDUCTION
The images are analysed to obtain first order
features such as mean, standard deviation, skewness,
kurtosis, entropy, energy and second order features such
as smoothness, IDM, contrast, correlation and
homogeneity. The first order values are calculated based
on the grey level histogram of the input images. These
histograms can help in extracting parameters with
statistics of different degrees.To offer a more precise
analysis, second order methods are applied. One of the
most well-known methods is grey-level co-occurrence
matrices (GLCM) to explain the second order features.
The first and second order features calculated for
analysis are described below.
• Mean: It represents the mean of grey levels
values. It is expressed as follows-


• Standard deviation: The measure of the
distribution of grey levels around the mean is
known as standard deviation.


• Skewness: The measure of dissimilarity of the
grey level histogram (if it is well balanced
around its mean or if it is more directed to the
left or to right with regard to its mean) is known
as skewness.


• Kurtosis: It is a measure that gives an indication
on the sharp and disentangled aspect or rather
flattened of histogram.
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
741

4. • Energy: It indicates how much intensity
variations is there in the image and is expressed
by-


• Entropy: It evaluates the complexity of the
image, i.e. how normal or non-normal the grey
level distribution is.


• Contrast: It is also called as sum of squares
variance. It is a measure of intensity of a pixel.

Generally, k = 2

• Local homogeneity: It is also known as inverse
difference moment (IDM). When the value of
GLCM is high near the main diagonal, it results
in high local homogeneity.
•


• Correlation: The spatial dependencies between
the pixels are predicted by the correlation
feature of the image.

A lot more features can be obtained from the image but
not all of them are useful. So principle component
analysis is used to reduce the features to the desired
level. A variance versus number of components graph is
plotted which is shown in section 4 of the paper.
C. SUPPORT VECTOR MACHINE
SVM is a supervised learning method which can
deal with the segmentation problem either as a
classification or regression problem. In this paper
segmentation is considered as a classification problem. It
works by generating a model file with the help of a
training dataset. The model file is then used to train the
classifier and validate the results with the test dataset.
The statistical features are first calculated from the
training set and then the exact same features are
calculated from the test dataset. The training dataset
features are fed to the classifier. The test dataset is then
used to validate the results. The process is repeated for
four different kernels, linear, RBF, polygonal and
quadratic.
D. GUI
The GUI is a modern tool in Matlab. It is an
interface file that can be used to retain and process
information for the GUI files created or modified in
Matlab [2]. The GUI layout editor can be used to
generate any kind of interface by dragging and placing
components in the blank GUI. The parameters will
depend upon the call back functions which are to be
attached along with the push buttons.
IV. EXPERIMENTS AND RESULTS
The dataset obtained from the The Cancer Imaging
Archive consists of 110 images including the normal
scans. Two groups of training and testing sets are formed
from the original dataset. 11 features including both first
and second order features are calculated from the
images. With the help of this algorithm, an interface has
been designed using the Matlab GUI program. The
variance versus number of components curve is shown in
figure 3 below.
10
0
10
1
10
2
0.4
0.5
0.6
0.7
0.8
0.9
1
No. of Principle Components
Var i anc e ( % )

Figure 3: Variance curve
As seen from figure 3, it can be noted that 20 principle
components are successful in sustaining almost 94% of
the total variance. Linear scale is used for y-axis while
log scale is used for x-axis.
Figure 4 and 5 shows the GUI result obtained
for benign and malignant tumor, respectively. Table 2
and 3 shows first order features obtained for benign and
malignant tumor, respectively. Table 4 and 5 shows
second order obtained for bening and malignant tumor,
respectively. Time evaluation has also been performed
for the whole analysis procedure. The time consumption
graph is shown in figure 6.
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
742

5. Figure 4: Segmented image for benign tumor obtained
using GUI

Figure 5: Segmented image for malignant tumor
obtained using GUI
Table 2: First order features for benign tumor
Image Mean Standard
Deviation
Entropy
1 0.0031 0.0897 3.1734
2 0.0041 0.0897 2.8045
3 0.0043 0.0897 3.0557
4 0.0034 0.0897 3.0432
5 0.0032 0.0897 3.0982
Image Kurtosis Skewness Energy
1 7.3281 0.4690 0.7621
2 17.3748 1.8112 0.7844
3 13.5271 1.1665 0.7876
4 9.6608 0.7404 0.7610
5 10.0859 0.9892 0.7667

Table 3: First order features for malignant tumor
Image Mean Standard
Deviation
Entropy
1 0.0041 0.0897 3.6530
2 0.0052 0.0896 3.1876
3 0.0054 0.0896 3.0267
4 0.0051 0.0896 3.3037
5 0.0061 0.0896 2.4981
Image Kurtosis Skewness Energy
1 6.7442 0.4810 0.7649
2 8.1742 0.7721 0.7488
3 12.4247 1.2966 0.7554
4 9.9701 0.8697 0.7553
5 205185 2.2369 0.8059

Table 4: Second order features for benign tumor
Image Smoothness IDM Contrast
1 0.9204 0.0576 0.2088
2 0.9398 0.3652 0.3501
3 0.9418 0.6349 0.3131
4 0.9272 0.0505 0.2566
5 0.9241 0.1109 0.3111
Image Correlation Homogeneity
1 0.1990 0.9351
2 0.1135 0.9372
3 0.0795 0.9397
4 0.1409 0.9315
5 0.0784 0.9338

Table 5: Second order features for malignant tumor
Image Smoothness IDM Contrast
1 0.9395 0.4858 0.2182
2 0.9514 0.3157 0.2750
3 0.9530 0.7951 0.3139
4 0.9507 0.3108 0.2794
5 0.9583 0.6852 0.3676
Image Correlation Homogeneity
1 0.1456 0.9352
2 0.1210 0.9290
3 0.0720 0.9293
4 0.0983 0.9302
5 0.1288 0.9428

From the tables above it can be seen that the second
order features outperform the first order features. The
first order features contains the information at individual
level while the second order features contains the
neighbourhood details of each pixel. Figure 8 shows the
time evaluation of the procedure.
1 2 3
0
0.005
0.01
0.015
0.02
0.025
Ta s k
Av er aged C omput at i on Ti m e ( s ec )

Figure 6: Time evaluation at different stages
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
743

6. The average time taken for feature extraction,
feature reduction and SVM classification is about 0.025
sec, 0.0146 sec and 0.0032 sec, respectively. The total
time taken in the entire evaluation process is 0.0428 sec.
From the graph it is seen that the most time consuming
task is feature extraction, it takes about 0.025 sec. On the
other hand, the least time consuming task is of SVM
classification taking only 0.0032 sec.
V. CONCLUSION
In this work, the MR images obtained are subjected
to classification using SVM and the algorithm is used to
develop an interface based on SVM. The GUI shows the
segmented image by just clicking on the push button.
Several first and second order features are calculated and
they are displayed on the GUI along with the four
kernels. The advantage of using GUI is that we can
modify the parameters according to our requirement
without having to rewrite the whole program. The
detection process becomes fast and efficient. The results
obtained show more accuracy and less time
consumption.

REFERENCES
[1] Xia, Y., Bettenger, K., Shen, L., Reis, A. “
Automatic segmentation of caudate nucleus from
human brain MR images”, IEEE Transactions and
medical imaging Vol. 26, No.4, 2007,pp. 509-517.
[2] E. Hassan and A. Aboshgifa, “Detecting Brain
Tumour from MRI Image using Matlab GUI
Program,” Int. J. Comput. Sci. Eng. Surv., vol. 6,
no. 6, 2015, pp. 47–60.
[3] Abd-Ellah, Mahmoud Khaled, Ali Ismail Awad,
Ashraf AM Khalaf and Hesham FA Hamed.
“Design and implementation of a Computer Aided
Diagnosis System for Brain Tumor
Classification”, in proc. 28th International
Conference on Microelectronics (ICM), Giza,
Egypt, 2016, pp.73-76.
[4] P. Jain, H. Didwania, and S. Chaturvedi, “Brain
Tumour extraction from MRI Images using
Matlab” 2nd Int. Conf. Emerg. Trends Eng.
Technol. Manag., pp. 282–289.
[5] M. Sudharson, S. R. T. Rajapandiyan, and P. U.
Ilavarasi, “Brain Tumor Detection by Image
Processing Using MATLAB,” Middle-East J. Sci.
Res., vol. 24, 2016, pp. 143–148.
[6] K. MohanaPriya, S. Kavitha, and B. Bharathi,
“Brain tumor types and grades classification
based on statistical feature set using support
vector machine,” in Proc. 10th International
Conference on Intelligent System and Control.
ISCO 2016, 2016,
pp. 1-8.
[7] Materka A and Strzelecki M, “Texture Analysis
Methods - A Review”, Technical University of
Lodz, Institute of Electronics, COST B11 Report,
Brussels, 1998, pp. 9-11.
[8] P. Kumar and B. Vijayakumar, “Brain Tumour Mr
Image Segmentation and Classification Using by
PCA and RBF Kernel Based Support Vector
Machine,” Middle-East J. Sci. Res., vol. 23, no. 9,
pp. 2106–2116, 2015.
[9] B. B. Shankaragowda, M. Siddappa and M.
Suresha, "A novel approach for the brain tumor
detection and classification using support vector
machine," in proc. 3rd International Conference
on Applied and Theoretical Computing and
Communication Technology (iCATccT), Tumkur,
2017, pp. 90-93.
[10] K. Machhale, H. B. Nandpuru, V. Kapur and L.
Kosta, "MRI brain cancer classification using
hybrid classifier (SVM-KNN)," in proc. 2015
International Conference on Industrial
Instrumentation and Control (ICIC), Pune, 2015,
pp. 60-65.
[11] N. Zhang, S. Ruan, S. Lebonvallet, Q. Liao, and
Y. Zhu, “Multi-kernel SVM based classification
for brain tumor segmentation of MRI multi-
sequence,” in Proc. Int. Conf. Image Process.
ICIP, pp. 3373–3376, 2009.
[12] M. F. B. Othman, N. B. Abdullah and N. F. B.
Kamal, "MRI brain classification using support
vector machine," 2011 Fourth International
Conference on Modeling, Simulation and Applied
Optimization, Kuala Lumpur, 2011, pp. 1-4.
[13] V. Wasule and P. Sonar, “Classification of brain
MRI using SVM and KNN classifier,” in Proc.
3rd IEEE Int. Conf. Sensing, Signal Process.
Secur. ICSSS 2017, pp. 218–223.
[14] The Cancer Imaging Archive (TCIA) Public
Access- http://wiki.cancerimagingarchive.net/




2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
744