optimization technique, to detect brain tumor in mri€¦ · optimization technique, to detect...
TRANSCRIPT
DOI: 10.23883/IJRTER.2017.3300.IMERL 234
Optimization Technique, To Detect Brain Tumor in MRI
Depika Patel1, Prof. Amit kumar Nandanwar
2
1Student, M.Tech, CSE, VNSIT Bhopal
2Computer Science andEngineering, VNSIT Bhopal
Abstract- Image Segmentation is an important and challenging factor in the field of medical
sciences. It is widely used for the detection of tumours. This paper deals with detection of brain
tumour from MR images of the brain. There are a variety of function of Image Segmentation among
them Medical Image Analysis is a popular application. Analysis of Medical Image provides in
sequence to the doctor for the treatment from MRI or CT images. Volumes of tissues, Brain Tumor
detection are some of the applications of image segmentation in medical image analysis. Many
researches have been occurred in order to detect tumor present in the brain, the area of tumor, the
type of tumor present. The following paper proposes a new algorithm for the detection of brain
tumor using MRI. The proposed method is implemented using Optimization Technique, to detect
tumor. The proposed method proves to be efficient, 16.39% accurate and 9.53% précised then the
existing work.
Keywords- MRI image, Brain Tumor, Genetic Algorithm, Curve Fitting and Support Vector
Machine.
I. INTRODUCTION
Human body is made up of several type of cells. Brain is a highly specialized and sensitive organ of
human body. Brain tumor is a very harmful disease for human being. The brain tumor is intracranial
mass made up by abnormal growth of tissue in the brain or around the brain. Brain tumour can be
detected by benign or malignant type. The benign being non-cancerous and malignant is cancerous.
Malignant tumour is classified into two types; primary and secondary tumour benign tumour is less
harmful than malignant. The malignant tumour it spread rapidly entering other tissues of the brain
therefore, worsening condition patients are loosed. Brain tumour detection is very challenging
problem due to complex structure of brain [1]Image Segmentation is a procedure that creates the
group of those pixels which has similar attributes. The results of image segmentation are used in
medical field for planning of treatment, recognition of tumor and analyzing the enlargement of tumor
[1]. Brain tumor is irregular and unrestrained growth of brain cells. Magnetic Resonance Imaging
provides minute information about brain tumor, its cellular structure, and it is an important tool for
diagnosis. According to Brain and Central Nervous System 372 people were diagnosed in TATA
Memorial Hospital in Mumbai, India. Out of 372, 250 were males (67%) and 122 were females
(33%). National Cancer Institute has approximate 22,070 novel cases of tumor and 12,920 deaths
due to brain tumor in the US for 2009. The Central Brain Tumor Registry of US (CBTRUS) has
made the survey which shows that 64,530 new cases of primary brain and central nervous system
tumors were diagnosed plough end of 2011. In United States in the year 2013 estimated new cases
are 23,130 and deaths are 14,080 from brain and other nervous system cancers [4].
In there time, there are numerous tools and software are available to detect and analyze tumor in the
brain. Since the last century has yielded much progression from which the chance of survival has
increased by diagnosing the tumor by means of the help of tools. But still there is a lot more left to
cure the brain tumor more efficiently, several researchers and authors have come forward in this field
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 235
and has designed several methods through which helps doctors in diagnosing and planning the
treatment [5].
There are several optimization technique in attendance which is being used in detecting tumor
efficiently and truthfully. because very last few years genetic algorithm has been in demand because
it has the advantage that it can extract features present in the image which is further used in
determining and identifying information present in the tumor. It is a searching algorithm based on the
process of natural evolution. Genetic Algorithm is used to generate the best solutions for
optimization and search problems. inherent algorithms are a subclass of evolutionary algorithms
which gives the best solutions to optimization problems using the techniques of natural evolution [2].
It is a technique used for feature selection and extraction. For the classification of features there are
several algorithms available. According to the training methods classifiers are divided to supervised
and unsubstantiated methods. The most popularly used classification method is Support Vector
Machine (SVM). It is a supervised learning method which classifies the several points into two
disjoint short spaces separated by a linear classifier [6]. This paper is prepared hooked on six
sections. Section II illustrates the work which has been done. Section III deals with a brief of the
methods used. Section IV explains the method designed to achieve the desired results. Section V
show the results obtained during the research. Finally, Section VI gives the termination of the
following paper.
The brain tumor detection can be done through MRI images. In image processing and image
enhancement tools are used for medical image processing to improve the quality of images. The
contrast adjustment and threshold techniques are used for highlighting the features of MRI images.
The Edge detection, Histogram, Segmentation and Morphological operations play a vital role for
classification and detecting the tumor of brain. The main objective of this paper is too studied and
reviewed the different research papers to find the various filters and segmentation techniques,
algorithms to brain tumor detection.
II. LITERATURE REVIEW Minakshi Sharma and Dr. Saurabh Mukherjee [1] in 2016 had compared the algorithms for
determining the tumor present in the brain. The comparison has been done among FCM, FCM+K-
Means, GA and ANFIS which gives improved accuracy, sensitivity and specificity. Amanpreet Kaur
and Gangandeep Jindal [2] in 2015 has given an approach through which tumor can be detected
effectively using Genetic Algorithm. It has been applied to reduce the population and then detecting
the tumor present in the brain. K. Selvanayaki and Dr. P.Kalugasalam [3] in 2013 have suggested an
algorithm for detecting tumor. The work has been done to compare the results of PSO, ACO and GA.
Meghana Nagori, Shivaji Mutkule and Praful Sonarkar [4] in 2013 has proposed a new approach of
detecting tumor in the brain. The comparison has been shown among various algorithms to achieve
accuracy in the detection of brain tumor. Dina Abdul Pahab, Samy S.A. Ghoniemy and Gamal
M.Selim [5] in 2012 have given an approach to detect tumor present in the brain by using
Probabilistic Neural Network Techniques. The PNN method is based on LVQ. Mehdi Jafari and
Reza Shafaghi [6] in 2012 have presented a method to detect brain tumor using GA and SVM. The
set of features has been extracted by GA and then these features have been classified using SVM.
Rabeb Mezgar, Ali Mahjoub. Mohamed, Salem. Randa asnd Mtibaa. Abdellatif [7] in 2012 has
proposed a method for detection of brain tumor by applying an algorithm of Expectation-
Maximization. A. Padma and R. Sukanesh [8] in 2011 has given an approach in which tumor has
been detected. The work shows that features have been extracted using Wavelet based and SGLDM
method. M. Guslen, A.E. Smith and D.M. Tate [9] has used genetic algorithm with curve fitting. The
research paper has used a sequential evolution mechanism which has been used to overcome
premature convergence to local minima when some terms are dominating others in magnitude. The
literature review shows that various methods have been employed to detect tumor in the brain. The
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 236
existing work has detected the tumor by segmenting the image by applying segmentation techniques.
But the rate of detection of tumor has been evaluated on the basis of sensitivity, specificity which is
not efficient for the detection and for the treatment of tumor. Hence, a new algorithm has been
proposed which will prove to be more efficient in detection of tumor and the performance will be
evaluated on the basis of sensitivity and specificity.
III. THEORY A. Genetic Algorithm:
A genetic algorithm is a search heuristic and iterative procedure. It involves a set of individuals
which is represented by a finite length of symbols, and encoded them into a possible solution in a
given problem. The procedure of genetic algorithm is as follows: the initial population is selected
randomly. At each evolutionary step which is called as generation, the individual in the selected
population is decoded according to some predefined parameters which is known as fitness function.
The new population is created by selecting population according to the fitness function obtained in
the previous iteration. The individual having high fitness function have the chance of reproducing
new population and those who have low fitness function are discarded. After then crossover is
applied to generate new offspring from the two selected individual called as parent who have best
fitness function by exchanging their properties. Then, the mutation is applied to prevent premature
convergence by selecting new points randomly. The bits are flipped with small probability PM. The
termination condition is specified by specifying fixed number of generation or achieving an
acceptable fitness level [3]
B. Sustains Vector Machine: Support Vector Machine is technique which is used for classification and has been proved that it has
better accuracy and computational advantages then conventional classification methods. The
technique performs training at the initial step which is based on predefining some features [6]. The
non-linear classification is performed with Kernel Trick. It is independent of the dimensions of the
feature space. It is a supervised learning technique and can be used for classification of n classes. It
can perform as a powerful tool by combining linear or non-linear functions. It has got its application
in medical imaging and data mining [8].
C. Curve Fitting: Curve Fitting is the process which approximates a function of closed form from a given data set. The
mathematical equations are used for analyzing and interpretation of data set to express it in closed
form. It is a preliminary step to several techniques which is used to solve and model problems. To
approach curve fitting there are several ways and methods available and few techniques are available
for complex functions as well [9].
IV.PROPOSED METHOD
The proposed work has been implemented using Genetic Algorithm, Curve Fitting and Support
Vector Machine. In this paper, Genetic Algorithm has been old to produce segment of the image.
The segments obtain with applying GA might be losing some information in neighboring segments.
To section image properly, with no the loss of information Curve Fitting has been used. Curve
Fitting has been used to get better the practice of segmenting. After segmenting the image, the
features have been extracted from the segments. These features are then classified using Support
Vector Machine. The classified data helps in determining the tumor of the brain using features which
have been extracted.
The proposed method has been explained with the assist of flow chart in Fig.1:-
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 237
Start
Input Image
Apply Genetic Algorithm
Apply Curve Fitting
Feature Extraction Repeat Algo for Better Result
Classification
Result
Stop
Fig.1 Flow Chart of Proposed Method
In this paper, initially training has been performed using Support Vector Machine on few sets of
tumor and non-tumor images as shown in Fig 2. The narrative like mean, median, range filter and
entropy has been calculated for tumor and non-tumor images which helps in classify among healthy
and non-healthy image. The resources then proceeds to section the input MRI brain image. The GA
has been applied to division the images by passing on the entropy as optimization parameters. Going
by the number of generations the Curve Fitting has been applied to get better the threshold which has
been derived using GA. The threshold determined for segmenting the MRI image strength behind
their information. because a result, Curve Fitting has been employed. On formative the threshold
obtained from Curve Fitting the MRI image segments. The features are extracted of the segments
approximating mean, median, standard deviation, texture, energy, edge, dissimilarity and entropy.
a. b. c. d.
Fig. 2 a) non- tumor image b) non- tumor image c) tumor image d) tumor image
The description extracted are classify according to the training performed. The categorization has
been performed by using sustain Vector Machine. The segments are classified on the basis of
features extracted. It results in detecting the tumor close by in the segment.
The proposed means is organism compared through the existing method in which Mahalanobis
Distance is used. It determines the detachment of two voxels [7].
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 238
V. EXPERIMENTAL RESULTS The proposed method begins by performing training on few sets of tumorous and non-tumorous
image as shown in Fig 2.
Fig. 3 Input MRI Image
Fig. 4 Best and Mean Fitness Value
The Fig.4 gives the Best and Mean Fitness Value as - 16.167 and 12.3222 acquire on pertain Genetic
Algorithm.
Fig.5 Curves obtained on Segmentation
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 239
The Fig.5 displays the curves obtained on segmenting the images into 5 segments
Fig. 6 Five Segments
The Fig.6 shows the five segments acquire following applying curve fitting.
Fig.7 Tumor Detected
The Fig.7 shows that finally tumor has been detected which has been shown by black curve.
TABLE 1.Parameters designed for obtaining segmentation
Parameters
Set
Values
Obtained
Values
Total Segments 5 5
Solution 0 0
Tolerance
Population Size 16 16
Maximum Generation 1000 1
Maximum Time(sec) 60 0.2990
International Journal of Recent Trends in Engineering & Research (IJRTER)
Volume 03, Issue 06; June - 2017 [ISSN: 2455-1457]
@IJRTER-2017, All Rights Reserved 240
It has explained in Table 1. that five segments are used with population size of 16 maximum
generation used is 1 out of 1000 in maximum time of 0.2990 sec with solution tolerance 0.
TABLE 2.Parameter values obtained when MRI applied 50 times
Parameters Mahalanobis Distance Proposed Method
Training Time(sec) 0.038982 0.056412
Matching Time 0.0128832 0.024135
True Positive 0.7846 0.9866
True Negative 0.9035 0.9798
False Positive 0.0965 0.0202
False Negative 0.2154 0.0134
Accuracy 0.8259 0.9898
Precision 0.8914 0.9867
Recall 0.7846 0.9866
F-measure 0.8156 0.9866
The Table 2. shows the parameters used to determine the efficiency of the proposed method with the
Mahalanobis Distance. It has been cleared from the table that the proposed method gives 16.39%
accuracy and 9.53% precision then the Mahalanobis Distance.
VI. CONCLUSION
In this paper, an algorithm has been proposed for detect the tumor. Genetic Algorithm, Curve Fitting
and Support Vector Machine has been employed to detect the image which shows the location where
tumor is present.
The proposed method has been compared with the previous method using Mahalanobis Distance on
the basis of few parameters on the Tumor images 50 times which has optimized the results. 16.39%
accuracy and 9.53% precision has been obtained on tumor images when applied 50 times. This result
shows that the proposed method proves to be highly beneficial for detection of tumor.
REFERENCES [1]. Minakshi Sharma, Dr. Saurabh Mukherjee, “Fuzzy C- Means, ANFIS and Genetic Algorithm for Segmenting
Astrocytoma –A Type of Brain Tumor” International Journal of Advanced Research in Computer Science and Software
Engineering, Vol. 3, Issue 6, June-2016, pp. 852-858.
[2]. Amarpreet Kaur, Gagandeep Jindal, “Overview of Tumor Detection using Genetic Algorithm”, IJIET, Vol. 2, Issue
2, April 2015.
[3]. K.Selvanayaki, Dr.P.Kalugasalam, “Intelligent Brain Tumor TissueSegmentation from Magnetic
Resonance Image (MRI) using MetaHeuristic Algorithms”, JGRCS, Vol. 4, February 2015.
[4]. Meghana Nagori, Shivaji Mutkule, Praful Sonarkar, “Detection of Brain Tumor by Mining MRI Images”,IJARCCE ,
Vol.2, Issue 4, January 2013.
[5]. Dina Aboul Dahab, Samy S. A. Ghoniemy, Gamal M. Selim,“Automated Brain Tumor Detection and
Identification Using Image Processing and Probabilistic Neural Network Techniques”, IJIPVC, Vol. 1, Issue 2, October
2012.
[6]. Mehdi Jafari, Reza Shafaghi, “A Hybird Approach for Automatic Tumor Detection of Brain MRI Using Support
Vector Machine and Genetic Algorithm”, GJSET, Issue 3, 2012, pp.1-8.
[7]. Mezgar. Rabeb, Ali Mahjoub. Mohamed, Salem. Randa and Mtibaa. Abdellatif, “Brain MRI Image
Segmentation in View Tumor Detection: Application to Multiple Sclerosis”, Springer, 2012, pp. 380-390.
[8]. A.Padma, R. Sukanesh, “Automatic Classification and Segmentation of Brain Tumor in CT Images using
Optimal Dominant Gray level Run length Texture Features”, IJACEA, Vol.2, No. 10, 2011.
[9]. Gulsen. M, Smith. A. E and Tate. D. M, “A genetic algorithm approach to curve fitting”, International Journal of
Production and Research, 1995.