International Journal of Computer Applications (0975 – 8887)

Volume 103 – No.1, October 2014

Brain Tumor Detection based on Machine

Learning Algorithms
Komal Sharma Akwinder Kaur Shruti Gujral
Research Scholar Assistant Professor Assistant Professor
M.E - CSE M.E - CSE M.E - CSE
CU, Gharuan CU, Gharuan CU, Gharuan
Mohali, Punjab, INDIA Mohali, Punjab, INDIA Mohali, Punjab, INDIA

ABSTRACT contains noise caused due to operator intervention which can

Automated defect detection in medical imaging has become lead to inaccurate classification. Large volume of MRI is to
the emergent field in several medical diagnostic applications. analyzed, thus, automated systems are needed as it they are
Automated detection of tumor in Magnetic Resonance more cost-effective. Automated detection of tumor in MRI
Imaging (MRI) is very crucial as it provides information images is necessary as high accuracy is needed when dealing
about abnormal tissues which is necessary for planning with human life.
treatment. The conventional method for defect detection in The MR human brain images are classified by using
magnetic resonance brain images is human inspection. This supervised techniques like artificial neural networks, support
method is impractical for large amount of data. So, automated vector machine, and unsupervised techniques like self-
tumor detection methods are developed as it would save organization map (SOM), fuzzy c-means when combined with
radiologist time. The MRI brain tumor detection is feature extraction techniques. Other supervised classification
complicated task due to complexity and variance of tumors. In techniques, such as k-nearest neighbors (k-NN) also group
this paper, tumor is detected in brain MRI using machine pixels based on their similarities in each feature [19].
learning algorithms. The proposed work is divided into three Classification of MR images either as normal or abnormal can
parts: preprocessing steps are applied on brain MRI images, be done via both supervised and unsupervised techniques.
texture features are extracted using Gray Level Co-occurrence
Matrix (GLCM) and then classification is done using machine In this paper, an efficient automated classification technique
learning algorithm. for brain MRI is proposed using machine learning algorithms.
The supervised machine learning algorithm is used for
Keywords: Magnetic Resonance Imaging, Segmentation, classification of brain MR image.
Feature Extraction, Texture Features, Machine learning.
2. RELATED WORK
1. INTRODUCTION Natarajan et al. [1] proposed brain tumor detection method for
The human body is composed of many types of cells. Each MRI brain images. The MRI brain images are first
cell has a specific function. The cells in the body grow. and preprocessed using median filter, then segmentation of image
divide in an orderly manner and form some new cells. These is done using threshold segmentation and morphological
new cells helps to keep the human body healthy and properly operations are applied and then finally, the tumor region is
working. When some cells lose their capability to control their obtained using image subtraction technique. This approach
growth, they grow without any order. The extra cells formed gives the exact shape of tumor in MRI brain image. Joshi et
form a mass of tissue which is called tumor. The tumors can al. [2] proposed brain tumor detection and classification
be benign or malignant. Malignant tumors lead to cancer system in MR images by first extracting the tumor portion
while benign tumors are not cancerous. According to a report from brain image, then extracting the texture features of the
published by central brain tumor registry of the United States detected tumor using Gray Level Co-occurrence Matrix
(CBTRUS), approximately 39,550 people were diagnosed (GLCM) and then classified using neuro-fuzzy classifier.
with benign and malignant brain tumors in 2002 . It indicates Amin and Mageed [3] proposed neural network and
that the rate of primary brain tumor whether malignant tumor segmentation base system to automatically detect the tumor in
or benign tumor is 14 per 100,000 [3]. brain MRI images. The Principal Component Analysis (PCA)
is used for feature extraction and then Multi-Layer Perceptron
The important factor in the medical diagnosis include the (MLP) is used classify the extracted features of MRI brain
medical image data obtained from various biomedical devices image. The average recognition rate is 88.2% and peak
that uses different imaging techniques like X-ray, CT Scan, recognition rate is 96.7%. Sapra et al. [4] proposed image
MRI. Magnetic Resonance Imaging (MRI) is a technique segmentation technique to detect brain tumor from MRI
which depends on the measurement of magnetic field vectors images and then Probabilistic Neural Network (PNN) is used
that are generated after an appropriate excitation of strong for automated brain tumor classification in MRI scans. PNN
magnetic fields and radio frequency pulses in the nuclei of system proposed handle the process of brain tumor
hydrogen atoms present in the water molecules of a patient's classification more accurately. Suchita and Lalit [5] proposed
body [5]. The MRI scan is much better than CT scan for unsupervised neural network learning technique for
diagnosis as it doesn't use any radiation. The radiologists can classification of brain MRI images. The MRI brain images are
evaluate the brain using MRI. The MRI technique can first preprocessed which include noise filtering, edge
determine the presence of tumor in the brain. The detection, then the tumor is extracted using segmentation. The
conventional method for tumor detection in MRI image is texture features are extracted using Gray-Level Co-occurrence
human inspection. This method is very time consuming. It is Matrix(GLCM) and then Self-Organizing Maps (SOM) are
not appropriate for large amount of data. The MRI also used to classify the brain as normal or abnormal brain, that is,

7
 International Journal of Computer Applications (0975 – 8887)
Volume 103 – No.1, October 2014

whether it contain tumor or not. Rajeshwari and Sharmila [6] Thus from extensive literature survey we found that most of
proposed pre-processing techniques which are used to the current brain tumor detection system uses texture,
improve the quality of MRI image before using it into an symmetry and intensity as features. Texture features are
application. The average, median and wiener filters are used important property of brain as texture perception has a very
for noise removal and interpolation based Discrete Wavelet important aspect in the human visual system of recognition
Transform (DWT) technique is used for resolution and interpretation [24]. Here, we propose extracting texture
enhancement. The Peak Signal to Noise Ratio (PSNR) is used features like energy, contrast, correlation, Homogenity [5].
for evaluation of these techniques. Gray Level Co-occurrence Matrix is used for extraction of
texture features.
George and Karnan [7] proposed MRI image enhancement
technique based on Histogram Equalization and Center Further we propose the use of ML algorithms to overcome the
Weighted Median (CWM) filter as they are used to enhance drawbacks of traditional classifiers. We investigate and
the MRI image more effectively. Daljit Singh et al. in [8] compare the performance of two machine learning algorithms
proposed a hybrid technique for automatic classification of namely MLP and Naive Bayes in this work. Since these ML
MRI images by first extracting the features using Principal algorithms are found to perform well in most of the pattern
Component Analysis (PCA) and Gray-Level Co-occurrence classification tasks. Neural networks are useful as they can
Matrix(GLCM) and then extracted features are fed as an input learn complex mappings between input and output. They are
to Support Vector Machine(SVM) classifier which classifies capable of solving much more complicated classification
the brain image as normal or abnormal. Gadpayleand and tasks. However, when certain rules cannot be modeled
Mahajani [9] proposed brain tumour detection and exactly, the concept of probability is used, which is the basis
classification system. The tumor is extracted using for Naive Bayes classification.
segmentation and then texture features are extracted using
GLCM and finally the BPNN and KNN classifiers are used to 3. PROPOSED METHOD
classify the MRI brain image into normal or abnormal brain. As per literature survey, it was found that automated brain
The accuracy is 70% using KNN classifier and 72.5% by tumor detection is very necessary as high accuracy is needed
using BPNN classifier. Shasidhar et al. in [10] proposed when human life is involved. Automated detection of tumor
modified Fuzzy C-Means (FCM) algorithm for MR brain in MR images involves feature extraction and classification
tumor detection. The texture features are extracted from brain using machine learning algorithm. In this paper, a system to
MR image and then modified FCM algorithm is used for brain automatically detect tumor in MR images is proposed as
tumor detection. The average speed-ups of as much as 80 shown in figure 1.
times a traditional FCM algorithm is obtained using the
modified FCM algorithm. The modified FCM algorithm is a
fast alternative to the traditional FCM technique. Rajesh and
Malar [11] proposed brain MR image classification based on
Rough set theory and feed-forward neural network classifier.
The features are extracted from MRI images using Rough set
theory. The selected features are fed as input to Feed Forward
Neural Network classifier which differentiate between normal
and abnormal brain and the accuracy of about 90% is
obtained.
Ramteke and Monali [14] proposed automatic classification
of brain MR images in two classes Normal and Abnormal
based on image features and automatic abnormality detection.
The Statistical texture feature set is obtained from normal
and abnormal images and then KNN classifier is used for
classifying image. The KNN obtain 80% classification rate.
Xuan and Liao [15] proposed statistical structure analysis
based tumor segmentation technique. The intensity-based,
symmetry-based and texture-based features are extracted from
MR image. Then, classification technique using AdaBoost is
used to classify the MR image into normal tissues and
abnormal images. The average accuracy of about 96.82% is
achieved. Othman et al. in [16] proposed Probabilistic neural
network technique for brain tumor classification. Firstly, the
features are extracted using the principal component analysis
(PCA) and the classification is performed using Probabilistic
Neural Network (PNN). Ibrahim et al. in [19] proposed
Neural Network technique for the classification of the
magnetic resonance human brain images. The features are
extracted using principal Component Analysis (PCA) and
then Back-Propagation Neural Network is used as a classifier
to classify MRI brain images as normal or abnormal. The
classification accuracy of about 96.33% is obtained. Jafari and
Shafaghi [21] proposed a hybrid approach for brain tumor
detection in MR images based on Support Vector
Machines(SVM). The texture and intensity features are used. Fig 1. Proposed Method for Brain Tumor Detection in MR
The accuracy of about 83.22% is achieved and is more robust. images

8
 International Journal of Computer Applications (0975 – 8887)
Volume 103 – No.1, October 2014

Image Acquisition: The MRI brain images are acquired and from the segmented image which shows the texture property
are given as input to pre-processing stage. The sample brain of the image. These features are extracted using Gray Level
MR images are shown in figure 2. Co-occurrence Matrix (GLCM) as it is robust method with
high performance. The GLCM texture feature extraction
method is very competitive as using smaller number of gray
levels shrinks the size of GLCM which reduces the
computational cost of the algorithm and at the same time
preserves the high classification rates. The GLCM features
are used to distinguish between normal and abnormal brain.
Texture contains some important information about surface
structural arrangement. The textural features based on gray-
tone spatial dependencies have a general applicability in
image classification
The GLCM texture features that are extracted are as follows:
Fig 2. Samples of brain MR image (1) Energy : It gives a measure of textural uniformity, that is,
measure of pixel pair repetitions.
Preprocessing: Preprocessing is needed as it provides
improvement in image data which enhances some of the ………(1)
image features which are important for further processing.
The pre-processing steps that are applied to MR image are as Range=[0,1]
follows :
(2) Contrast : It gives a measure of intensity contrast between
The RGB MR image is converted to gray scale image and a pixel and its neighbor over the whole image.
then median filter is applied for noise removal from brain MR
images as shown in figure 3(b). The noise is to removed for …….(2)
further processing as high accuracy is needed.
Range=[0,1]
Then edges are detected from filtered image using canny edge
detection as shown in figure 3(c). The edge detected image is (3) Correlation : It gives a measure of how correlated a pixel
needed for segmentation of the image. to its neighbor over the whole image.
Then watershed segmentation is done for finding the location . .(3)
of the tumor in the brain image as shown in figure 3(d).
Segmentation is the process of dividing an image into Range=[-1,1]
multiple segments. The aim of segmentation is to change
representation of image into something which is more easy to (4) Homogeneity : It gives a measure of closeness of
analyze. The result of watershed segmentation is label image. distribution of elements in GLCM to GLCM diagonal.
In label image, all the different objects identified will have
…..(4)
different pixel values , all the pixels of first object will have
value 1, all the pixels of second object will have value 2 and Range=[0,1]
so on [23]. The various preprocessing operations applied on
brain MR image are shown in figure 3. Thus to classify the brain as normal or abnormal , the MRI
brain images are collected, preprocessing steps and
segmentation are then applied. The segmented image is used
for feature extraction and texture features are extracted using
GLCM. The preprocessing and feature extraction steps are
done in Matlab 2012a.
(5) Classification: The Machine learning algorithms are used
for classification of MR brain image either as normal or
abnormal. The major aim of ML algorithms is to
automatically learn and make intelligent decisions. The
feature set formed by above specified method was applied to
Multi-Layer Perceptron (MLP) and Naive Bayes for
classification. MLP [3] is a feed forward artificial neural
network model that maps sets of input data into a set of
appropriate output. It is known as feed forward because it
does not contain any cycles and network output depends only
on the current input instance. In MLP, each node is a neuron
Fig 3(a-d). Preprocessing operations on input brain image with a nonlinear activation function. It is based on supervised
learning technique. Learning take place by changing
Feature Extraction: When input to an algorithm is very large
connection weights after each piece of data is processed,
and redundant to be processed, it is transformed into reduced
based on the amount of error in the target output as compared
representative set of features called feature vector.
to the expected result. The goal of the learning procedure is to
Transformation of input data into set of features is called
minimize error by improving the current values of the weight
feature extraction [5]. In this step, the important features
associated with each edge. Because of this backward changing
needed for image classification are extracted. The segmented
process of the weights, model is named as back- propagation.
brain MR image is used and texture features are extracted

9
 International Journal of Computer Applications (0975 – 8887)
Volume 103 – No.1, October 2014

Naive bayes is a supervised learning as well as statistical

method for classification. It is simple probabilistic classifier
70

Time Taken (in sec)

based on Bayes theorem. It assumes that the value of a 60
particular feature is unrelated to the presence or absence of 50
any other feature. The prior probability and likelihood are
calculated in order to calculate the posterior probability. The 40
method of maximum posterior probability is used for 30
parameter estimation. This method requires only a small 20
amount of training data to estimate the parameters which are
needed for classification. The time taken for training and 10
classification is less. 0
MLP Naïve bayes
4. EXPERIMENTAL RESULTS
The experiment was carried out on 212 brain MR images. ML Algorithms
From each image, the texture based features are extracted and
weka tool is used for classification [28]. The texture based
features such as energy, contrast, correlation, homogeneity are
extracted using GLCM. The Multi-Layer Perceptron (MLP) Fig 5. Graphical representation of time taken
and Naïve bayes with 66% percentage split is used for
classification. In 66% percentage split, 66% of the instances
are used for training and remaining instances are used for The MLP gives more accuracy and takes more time to build
testing. the model while the time taken by Naïve Bayes is less and is
Table 1. Experimental result analysis less accurate. Due to diverse appearances and complexity of
tumors, the proposed method gives the satisfied accuracy. The
ML Total Model Build Classification high accuracy is desirable as human life is involved.
Algorithm samples Time Rate (%)
5. CONCLUSION
MLP 210 61.82 98.6 This paper proposes a work on brain tumor detection system
based on machine learning algorithms. The texture based
Naive 210 0.02 97.6 features are extracted using Gray Level Co-occurrence Matrix
bayes (GLCM). The texture features of the image considered in this
proposed work include energy, contrast, correlation,
homogeneity. For the classification purpose, Multi-Layer
From the Table 1, we can find the classification rate of brain Perceptron and Naïve bayes machine learning algorithm is
MR images using MLP and Naive bayes. The accuracy of used and the maximum accuracy 98.6% and 91.6% is
about 98.6% and 91.6% is obtained respectively. achieved by considering 212 samples of brain MR images.
This accuracy can probably be increased by considering a
100 large data set and extracting intensity based features in
addition to the texture based features.
98
Accuracy in %

96 6. REFERENCES
[1] Natarajan P, Krishnan.N, Natasha Sandeep Kenkre,
94 Shraiya Nancy, Bhuvanesh Pratap Singh, "Tumor
Detection using threshold operation in MRI Brain
92
Images" , IEEE International Conference on
90 Computational Intelligence and Computing Research,
2012.
88
MLP Naïve bayes [2] Dipali M. Joshi, N. K. Rana, V. M. Misra, "
Classification of Brain Cancer Using Artificial Neural
ML Algorithms Network" , IEEE International Conference on Electronic
Computer Technology ,ICECT ,2010.
[3] Safaa E.Amin, M.A. Mageed," Brain Tumor Diagnosis
Fig 4. Graphical representation of accuracy Systems Based on Artificial Neural Networks and
Segmentation Using MRI" , IEEE International
Conference on Informatics and Systems, INFOS 2012.
[4] Pankaj Sapra, Rupinderpal Singh, Shivani Khurana,
"Brain Tumor Detection Using Neural Network" ,
International Journal of Science and Modern
Engineering, IJISME ,ISSN: 2319-6386, Volume-1,
Issue-9, August 2013.
[5] Suchita Goswami, Lalit Kumar P. Bhaiya, " Brain Tumor
Detection Using Unsupervised Learning based Neural
Network" , IEEE International Conference on

10
 International Journal of Computer Applications (0975 – 8887)
Volume 103 – No.1, October 2014

Communication Systems and Network [17] Shweta Jain, "Brain Cancer Classification Using GLCM
Technologies,2013. Based Feature Extraction in Artificial Neural Network" ,
International Journal of Computer Science &
[6] S. Rajeshwari, T. Sree Sharmila, "Efficient Quality Engineering Technology ,IJCSET, ISSN : 2229-3345
Analysis of MRI Image Using Preprocessing Vol. 4 No. 07 Jul 2013.
Techniques", IEEE Conference on Information and
Communication Technologies, ICT 2013. [18] Dina Aboul Dahab, Samy S. A. Ghoniemy, Gamal M.
Selim, "Automated Brain Tumor Detection and
[7] E. Ben George, M.Karnan, "MRI Brain Image Identification using Image Processing and Probabilistic
Enhancement Using Filtering Techniques", International Neural Network Techniques" ,International Journal of
Journal of Computer Science & Engineering Image Processing and Visual Communication, ISSN
Technology, IJCSET, 2012. 2319-1724 : Volume (Online) 1 , Issue 2 , October 2012.
[8] Daljit Singh, Kamaljeet Kaur, "Classification of [19] Walaa Hussein Ibrahim, Ahmed Abdel Rhman Ahmed
Abnormalities in Brain MRI Images Using GLCM, PCA Osman, Yusra Ibrahim Mohamed, "MRI Brain Image
and SVM" , International Journal of Engineering and Classification Using Neural Networks" ,IEEE
Advanced Technology (IJEAT) ISSN: 2249 – 8958, International Conference On Computing, Electrical and
Volume-1, Issue-6, August 2012. Electronics Engineering, ICCEEE,2013.
[9] Prachi Gadpayleand, P.S. Mahajani, "Detection and [20] Noramalina Abdullah, Lee Wee Chuen, Umi Kalthum
Classification of Brain Tumor in MRI Images ", Ngah Khairul Azman Ahmad, "Improvement of MRI
International Journal of Emerging Trends in Electrical Brain Classification using Principal Component
and Electronics, IJETEE – ISSN: 2320-9569, Vol. 5, Analysis" , IEEE International Conference on Control
Issue. 1, July-2013. System, Computing and Engineering, 2011.
[10] M. Shasidhar , V.Sudheer Raja, B. Vijay Kumar, "MRI [21] Mehdi Jafari, Reza Shafaghi, "A Hybrid Approach for
Brain Image Segmentation Using Modified Fuzzy C- Automatic Tumor Detection of Brain MRI using Support
Means Clustering Algorithm" ,IEEE International Vector Machine and Genetic Algorithm", Global Journal
Conference on Communication Systems and Network of Science Engineering and Technology, Issue-3, 2012.
Technologies, 2011.
[22] V. Salai Selvam and S. Shenbagadevi, "Brain Tumor
[11] T. Rajesh, R. Suja Mani Malar,” Rough Set Theory and Detection using Scalp EEG with Modified Wavelet-ICA
Feed Forward Neural Network Based Brain Tumor and Multi Layer Feed Forward Neural Network" ,
Detection in Magnetic Resonance Images" ,IEEE Annual International Conference of the IEEE EMBS
International on Advanced Nanomaterials & Emerging Boston, Massachusetts USA, August 30 - September 3,
Engineering Technologies, 20 13. 2011.
[12] Komal Sharma, Navneet Kaur, " Comparative Analysis [23] Amir Shahzad, Muhammad Sharif, Mudassar Raza,
of Various Edge Detection Techniques" , International Khalid Hussai, " Enhanced Watershed Image Processing
Journal of Advanced Research in Computer Science and Segmentation " , Journal of Information &
Software Engineering, IJARCSSE, ISSN: 2277 128X, Communication Technology, Vol. 2, No. 1, 2009.
Volume 3, Issue 12, December 2013.
[24] Neelam Marshkole, Bikesh Kumar Singh, A.S Thoke,
[13] J. Selvakumar, A. Lakshmi, T. Arivoli, " Brain Tumor "Texture and Shape based Classification of Brain
Segmentation and Its Area Calculation in Brain MR Tumors using Linear Vector Quantization" , International
Images using K-Mean Clustering and Fuzzy C-Mean Journal of Computer Applications (0975 – 8887) Volume
Algorithm" , IEEE-International Conference On 30– No.11, September 2011.
Advances In Engineering, Science And Management,
ICAESM, 2012. [25] MATLAB Central, URL:
http://www.mathworks.in/discovery/imagesegmentation.
[14] R. J. Ramteke1, Khachane Monali Y., " Automatic html.
Medical Image Classification and Abnormality Detection
Using K-Nearest Neighbour" , International Journal of [26] MATLAB Central, URL:
Advanced Computer Research,Volume-2 Number-4 http://www.mathworks.in/matlabcentral/fileexchange/22
Issue-6 December-2012. 187-glcm texture-features.
[15] Xiao Xuan, Qingmin Liao, Statistical Structure Analysis [27] MATLAB Central, URL:
in MRI Brain Tumor Segmentation" ,IEEE International http://www.mathworks.in/help/images/analyzing-the-
Conference on Image and Graphics, 2007. texture-of-an-image.html.
[16] Mohd Fauzi Othman, Mohd Ariffanan, Mohd Basri, " [28] WEKA 3: Data Mining With Open Source Machine
Probabilistic Neural Network for Brain Tumor Learning Software in JAVA, URL:
Classification" ,IEEE International Conference on http://www.cs.waikato.ac.nz/ml/weka/.
Intelligent Systems, Modelling and Simulation,2011.

IJCATM : www.ijcaonline.org 11