DATE OF ACCEPTANCE: APRIL 18, 2014 ISSN: 2348-4098

DATE OF PUBLICATION: APRIL 29, 2014 VOLUME 02 ISSUE 04 APRIL-MAY 2014

DETECTIONANDCLASSIFICATIONOFBRAINTUMORSINMRIIMAGES

1
ABHIJEETZAMRE,2TANVEERSHAH,3HITESHTHADHANI,4DHANANJAYBANGAR

1,2,3,4
ELECTRICALENGINEERINGDEPARTMENT,VJTI,MAHARASHTRA,INDIA

ABSTRACT

Inthispaper,detectionofbraintumorisdoneusingMRIimages.Braintumorisanabnormal
intracranial growth caused by cells reproducing themselves in an uncontrolled manner.
Curing cancer has been a major goal of medical researchers for decades. The early
detection of cancer can be helpful in curing the disease completely. In this paper we
proposeanANNbaseapproachtoidentifybraintumorfromMRIimages.Thealgorithm
incorporatesstepsforpreprocessing,imagesegmentation,featureextractionandimage
classification using neural network techniques. A classification is done into two classs i.e
brainintohealthybrainorabrainhavingatumor.Evaluationwasperformedonimagedata
baseof25BrainTumorimages.Theproposedmethodgivesfastandbetterrecognition
rate,finallythetumorareaisspecifiedbyregionofinteresttechniqueasconfirmationstep.A
userfriendlyMATLABGUIprogramhasbeenconstructedtotesttheproposedalgorithm.

IndexTerms:braintumor;imagesegmentation;featureextraction;neuralnetworks

1. INTRODUCTION five years, making it the most fatal

In the last decades, we have been cancer among children. Its associated

observed a dynamic growth in the withneurologicaldisabilities,retardation

number ofresearch works conducted in and psychological problems and

the region of cancer diagnosis. Many increased risk of death. Magnetic

universitycenters[1]arefocusedonthe resonanceimaging(MRI)isconsiderednow

issue because of the fact that cerebral asanimportanttoolforsurgeons.Abrain

cancer is spreading among the world tumor is any intracranial mass created by

population. For example in the US, abnormal and uncontrolled cell division.

nearly3000childrenarediagnosedwith Magnetic resonance imaging (MRI) or

braintumors.Almosthalfwilldiewithin magnetic resonance tomography (MRT)

INTERNATIONAL JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in 46

DATE OF ACCEPTANCE: APRIL 18, 2014
DATE OF PUBLICATION: APRIL 29, 2014
is a medical imaging technique used in
radiology to investigate the anatomy
and function of the body in both health
and disease. MRI scanners use strong
magneticfieldsandradiowavestoform
images of the body. The technique is
widely used in hospitals for medical
diagnosis, staging of disease and for
followup without exposure to ionizing
radiation.Ifthelocationofabraintumor
is accessible for surgery, the
neurosurgeon may remove as much as
possible of the tumor. The paper is
organizedasfollows.InSection1,image
enhancement techniques are applied to
remove noise. In Section 2, Image
segmentation techniques are applied.
Feature extraction methods used is
explainedinsection3.Insection4,neural
network classifiers are employed to
classifyweathertheMRIimageisnormal
or consisting of brain tumor and
classification is done using radial basis
function based Linear Regression
Algorithm.
INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in
ISSN: 2348-4098
VOLUME 02 ISSUE 04 APRIL-MAY 2014
1.1 IMAGEACQUISITION
The MRI images of brain tumor are
collected from internet. The images are
converted into standard size of 200 x
200. The images are converted to gray
scaleimagesforfurtherprocessing.
1.2IMAGEENHANCEMENTTECHNIQUES
It refers to accentuation, or sharpening,
ofimagefeaturessuchasboundaries,or
contrasttomakeagraphicdisplaymore
usefulfordisplay&analysis.Thisprocess
does not increase the inherent
information content in data. It includes
graylevel&contrastmanipulation,noise
reduction,filtering.Imageenhancement
isdonebyfollowingmethods.
Image Filtering: In Image filtering
Gaussian Filter is used and laplacian of
gaussianisobtained.Intwodimensions,
itistheproductoftwoonedimensional
Gaussians,oneperdirection:
wherexisthedistancefromtheoriginin
thehorizontalaxis,yisthedistancefrom
theoriginintheverticalaxis,andisthe
standard deviation of the Gaussian
distribution. The Laplacian L(x,y) of an
47
 DATE OF ACCEPTANCE: APRIL 18, 2014
DATE OF PUBLICATION: APRIL 29, 2014
imagewithpixelintensityvaluesI(x,y)is
givenby:
Laplacian filters are derivative filters
used to find areas of rapid change
(edges)inimages.Sincederivativefilters
areverysensitivetonoise,itiscommon
to smooth the image (e.g., using a
Gaussian filter) before applying the
Laplacian. This twostep process is call
the Laplacian of Gaussian (LoG)
operation.When the image is uniform,
the LoG will give zero. Wherever a
change occurs, the LoG will give a
positiveresponseonthedarkersideand
a negative response on the lighter side.
At a sharp edge between two regions,
theresponsewillbe
positivejusttooneside
zeroawayfromtheedge
negativejusttotheotherside
zero at some point in between on the edge
itself
UnsharpMasking:The"unsharp" ofthe
name derives from the fact that the
technique uses a blurred, or "unsharp",
first image is blurred and subtracted
from original image so edges can be
INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in
ISSN: 2348-4098
VOLUME 02 ISSUE 04 APRIL-MAY 2014
obtained and then these edges are
added to original image itself so
resultantimageisgoodincontrast.
Thissharpenimageisnowerodedusing
morphological filter. This image is very
smooth. We convert this image into
binary image with a threshold of 0.6.
Thuswegetablackandwhiteimagein
whichwhitepartindicatestumor.
2. IMAGESEGMENTATION
Image segmentation is the process of
assigning a label to every pixel in an
image such that pixels with the same
label share certain visual characteristics.
Theresultofimagesegmentationisaset
of segments that collectively cover the
entire image, or a set of contours
extracted from the image (see edge
detection).Eachofthepixelsinaregion
are similar with respect to some
characteristic or computed property,
suchascolor,intensity,ortexture.
Thus, there was no need to investigate
more sophisticated image segmentation
algorithms. The Canny Edge detection
technique is used. The goal of edge
detectionistomarkthepointsinadigital
image at which the luminous intensity
changes sharply. Sharp changes in image
propertiesusuallyreflectimportantevents
48
DATE OF ACCEPTANCE: APRIL 18, 2014
DATE OF PUBLICATION: APRIL 29, 2014
Canny Edge detection proceeds thorough
followingsteps:
Smooth the image with a Gaussian
filter,
Compute the gradient magnitude
and orientation using finitedifference
approximationsforthepartialderivatives,
Apply nonmaxima suppression,
withtheaidoftheorientationimage,to
thinthegradientmagnitudeedge
Trackalongedgesstartingfromany
point exceeding a higher threshold as
longastheedgepointexceedsthelower
threshold.
Applyedgelinkingtofillsmallgaps
3. FEATUREEXTRACTION
In this section, we used PCA to extract
features. First, PCA was performed on
themetaboliteconcentrationdatasetto
extract feature vectors. We obtained all
eigen values and their corresponding
eigen vectors for image. The eigen
Vector corresponding to the highest
eigenvalueisusedasfeaturevector.The
length of vector is 200. For convience 5
elementsareshownhere.
0.00328 0.00114 0.0013 0.00334
INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in
ISSN: 2348-4098
VOLUME 02 ISSUE 04 APRIL-MAY 2014
Anewdatasetwasreconstructedforall
the normal healthy brain images and
braintumorimages.
4. CLASSIFICATION USING
RADIALBASISFUNCTION
In Neural networks based classification;
training of neural networks is done to
perform the required classification task
using different input features (Feature
vectors obtained using Eigen Values).
The neural network is used to classify
brains into healthy brain or brain with
tumor.ThenumberofNeuronsusedare
20andwhichgivesaMeanSquareError
of 0.012. As the number of Neurons
increases.MeanSquareErrordecreases
and the output is close to the desired
one and reduces the chances of
misclassification increasing the accuracy
ofcorrectclassification.
Radial basis networks can be used to
approximate functions.Itadds neurons
to the hidden layer of a radial basis
network until it meets the specified
mean squared error goal. The
largerspreadis, the smoother the
function approximation. Too large a
spread means a lot of neurons are
0.00034 required to fit a fastchanging function.
49
DATE OF ACCEPTANCE: APRIL 18, 2014 ISSN: 2348-4098
DATE OF PUBLICATION: APRIL 29, 2014 VOLUME 02 ISSUE 04 APRIL-MAY 2014

Toosmallaspreadmeansmanyneurons
are required to fit a smooth function,
and the network might not generalize
well. Commonly used types of radial
basisfunctionsinclude:

Fig1:inputimagewithLoGandunsharp

Radial basis functions are typically used masking

to build upfunction approximationsof

theform Morphologicaloperationsperformedusing
BallStructuringelement.


wheretheapproximatingfunctiony(x)is
represented as a sum ofNradial basis
functions, each associated with a
different centerxi, and weighted by an
appropriate coefficientwi. The Fig2:morphologicalerodedimage
weightswican be estimated using the
matrix methods oflinear least squares, The above image is threshold to binary
because the approximating function usingthresholdvalue0.6andisshownin
islinearintheweights. fig3

5. CONCLUSION

Input image is taken and preprocessing

is done on it. For image enhancement

LOGisusedalongwithUnsharpMasking

giving an output image as shown in Fig3:Binaryimage

figure1.

Canny edge detector is applied on the

black and white segmented image and

INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in 50
DATE OF ACCEPTANCE: APRIL 18, 2014 ISSN: 2348-4098
DATE OF PUBLICATION: APRIL 29, 2014 VOLUME 02 ISSUE 04 APRIL-MAY 2014

thetumorisdetected(outerboundaryof & Characterization of Brain Tumors",.

tumor)asshowninfig4. 9781467312288/12/2012IEEE.
[3]NeilM.Borden,MD,ScottE.Forseen,
MD, "Pattern Recognition
Neuroradiology", Cambridge University
Press,NewYork,2011.
[4] Noramalina Abdullah, Umi Kalthum
Ngah, Shalihatun Azlin Aziz" Image
Fig4:segmentedimage ClassificationofBrainMRIUsingSupport
Vector Machine", 978161284896

6. CLASSIFICATION 9/11/2011IEEE.
[5]M.J.Er,W.Chen,S.Wu,"Highspeed

space recognition based on discrete

CORRECTLY INCORRECTLY
cosine trans form and RBF Neural
CLASSIFIED CLASSIFIED
Network". IEEE Trans. On Neural
TRAIN 25 0
INPUTS network,Vol,16,NoJ,pp679,691,2005.
TESTINPUTS 5 3 [6] Z.Pan, AG. Rust, H. Bolouri" "Image

redundancy reduction for neural

network classification using discrete
ACKNOWLEDGEMENT
cosine trans form", Prop. Int. Conf on
Dr.MeenaPhanse,HeadofDepartment,
Neural Network,VoI.3,pp 149,154,
Electrical Engineering Department,
Itali,2000.
VeermataJijabaiTechnologicalInstitute,
[7] AK. Jain, R.P. W. Duin, J. Mao,
Mumbai,India400019
Statistical pattern recognition: a review,

REFERENCES
[I] Issac H. Bankman, Hand Book of
Medical Image processing and Analysis,
AcademicPress,2009.
[2] Sarah Parisot, Hugues Duffau, sr
ephane Chemouny, Nikos Paragios,
"Graphbased Detection, Segmentation
IEEETrans.PatternAnal.Mach.Iintell.22
(2000)437.
[8] K. Fukunaga, Introduction to
Statistical pattern recognition, 11 nd
Edition. Academic Press, New York,
1990.
[9] N. Kwak, and C.H. Choi, "Input
Feature Selection for classification

INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in 51
DATE OF ACCEPTANCE: APRIL 18, 2014 ISSN: 2348-4098
DATE OF PUBLICATION: APRIL 29, 2014 VOLUME 02 ISSUE 04 APRIL-MAY 2014

problem", IEEE Transactions on Neural BIOGRAPHIES

Networks,13(1),143159,2002.
[10] D.F. Specht, "Probabilistic Neural
Abhijeet Zamre,
Networks for classification, mapping, or
Student, Electrical
associative memory", Proceedings oj
department,VJTI
IEEE International Conjerence on Neural
Networks, Vol I, IEEE Press, New York,
pp.525532,june1988.
[11] D.F. Specht, "Probabilistic Neural Tanveer shah,

Networks" Neural Networks, Vol. 3, No Student, Electrical

l,pp.109118,1990. department,VJTI

[12] Mohd Fauzi Othman and Mohd

Ariffanan Mohd Basri, "Probabilistic
Neural Network For Brain Tumor
Classification", Second International Hitesh Thadhani,
Conference on Intelligent Systems, Student, Electrical
ModelingandSimulation,IEEEComputer department,VJTI
Society,2011,00110.1109/ISMS.2011J2.
[13] Li, H. Y., Fan, W. Y., 2008. Matlab
Realization of Sensing Image
Classification Based on Probabilistic Dhananjay Bangar,
Neural Network. journal oj Northeast Student, Electrical
ForestryUniversity,36(6),pp.6466. department,VJTI

INTERNATIONAL
JOURNAL OF SCIENCE, ENGINEERING AND TECHNOLOGY- www.ijset.in 52