IRJET- Brain Tumor Detection and Classification with Feed Forward Back Propagation Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 7367
Brain Tumor Detection and Classification with Feed Forward Back
propagation Network
Neha K Mohanan1, Akhila Gopal2, Ayana Ajith3, Aswathy M.R4
1,2PG Scholar, Department of Computer Science & Engineering, Vidya Academy of Science & Technology , Kerala,
India
3,4Asst. Professor, Department of Computer Science & Engineering, Vidya Academy of Science & Technology,
Kerala, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Brain is an organ that controls activities of all
the parts of the body. Recognitionofautomatedbraintumorin
Magnetic resonance imaging (MRI) is a difficult task due to
complexity of size and location variability. Previous methods
for tumor are time consuming and less accurate. Detection of
tumor can be done by MRI and CT scan. MRI give high quality
images of the body parts and is often used while treating
tumors. There are many methods to building completely
mechanized computer aided diagnosis (CAD) framework to
help therapeutic experts in recognizing and diagnosing brain
tumor. Different stages in brain tumor detection are Image
Acquisition, Image Preprocessing, Feature Extraction, Image
Segmentation and Classification. Previous methods fortumor
detection are time consuming and less accurate. Inthepresent
work, Recognize and model nonlinear relationships between
data. In preprocessing, divides images into blocks using
bounding box method. So we get large number of datasetfrom
each image. Then Instead of a single image here every blocks
of a image handled. Then Haar transform used to noise
removal and smoothing with preserving edge information.
Statistical analysis, GLCM, Morphologicaloperationsandedge
are used for feature extraction. Extracting featuresfromeach
blocks of an image. This method helps to identifyingabnormal
areas from images. Then Block matching techniquewithFFBP
used for classification. Feed forward Back propagationneural
network is a multilevel error feed-back network that reduces
error rate. Well design and training of ANN make it qualified
for decision making operations when it faced with new data.
This method results highaccuracyandlessiterationsdetection
which further reduces the consumption time. This automatic
segmentation algorithm gives shape, size and location of the
tumor more accurately. Classifying brainMRI into normaland
abnormal cases. We used a benchmark dataset MRI brain
images. The experimental results show that our approach
achieves 98% classification accuracy using ANN.
Key Words: MRI; Brain tumor; feature extraction; Feed
Forward Back propagation network.
1. INTRODUCTION
Brain tumor is only any mass that outcomes from an
anomalous and an uncontrolled growth of cells in the brain.
Brain tumor is progressively reparable and treatable
whenever distinguished at beginning period; it can increase
the intracranial pressure which can spoil the brain
permanently. Brain tumor symptoms depend upon the size
of tumor, location and its type. Its threat levels depend upon
the combination of factors like thetypeoftumor,itsposition,
its size and its state of growth. Brain tumors can be
malignant (malignant) or non- malignant (benign). Benign
brain tumors are a low grade and it is said to be non-
cancerous brain tumors, which grows slowly and pushaside
normal tissue but do not invade the surrounding normal
tissue. They are homogeneous, well defined and are known
as non- metastatic tumors, as they do not form any
secondary tumor. The malignantbraintumorsarecancerous
brain tumors, which grows rapidly and invade the
surrounding normal tissue. Malignant brain tumors or
cancerous brain tumors can be counted among the at most
deadly diseases. Detection of tumor can be done by MRI and
CT scan. Sometimes cancer diagnosis can be delayed or
missed because of some symptoms. Theprincipleaimofthis
paper is to analyze the best segmented method and classify
them for a better performance. Here Feed-forward back
propagation neural system is utilized to characterize the
execution of tumors part of the image. Artificial Neural
Networks (ANNs) consist of an interconnection of simple
components referred to as neurons,whichareprogramming
constructs that mimic the properties of biological neurons.
ANNs consist of one or more layers. Each layer has one or
more neurons. The neuron (perceptron) can be defined
simply as a device with many inputs, one output, and an
activation function. This strategy results high exactness and
less cycles recognition which additionally diminishes the
utilization time.
1.1 MRI
MRI makes a revolution in the medical field. Magnetic
Resonance Imaging (MRI) is a restorative imaging strategy
used to envision the interior structure of the body and
provide high quality images. MRI provides a greater
distinguishing between the different tissues of the body. A
magnetic resonance imaging instrument or MRI Scanner
uses powerful magnets to polarize and excite hydrogen
nuclei i.e. proton in water molecules in human tissue,
producing a detectable signal which is spatially encoded,
resulting in images of the body MRI contains useful and fine

information that can be used in improving the quality of
diagnosis and treatment of brain. MRI image texture
contains a rich source of information such as characterize
brightness, color, slope, size, and other features. Alongthese
lines, it is required to figure the computerized image
qualities to depict its texture propertiesnumericallyandthis
is called image feature extraction. Feature extraction are
distinguishing applicable highlights prompts quicker,
simpler, and better to get images. Featureextractionprocess
influences essentially the natureoftheclassificationprocess.
1.2 Automated System
Automated system(detection) of braintumorthroughMRIis
basically called Computer Aided Diagnosis (CAD) system.
The CAD system can provide highly accurate reconstruction
of the original image i.e. the valuableoutlook andaccuracyof
earlier brain tumor detection. It consists of two or more
stage. In the initial stage preprocessing has required after
that stages post-processing i.e. segmentation are required.
Then detection strategies and other information, feature
extraction, feature selection, classification,andperformance
analysis are compared and studied. Pre-processing
techniques are used to improvement of image quality and
remove noise for the accurate detection of the undesired
regions in MRI. Post-processing is used to segment with
different strategy the brain tumor from the MRI of brain
images.
2. RELATED WORK
In[1], the authors proposed Classification of MRI Brain
Images using K-NN and k-means. The diagnosis based on the
classification of magnetic resonance images(MRI).The
Method consists of three stages. In the first stage the imageis
filtered using Discrete Wavelet Transform(DWT )and the
features are extracted using Principal component
analysis(PCA).In the second stage the segmentation of MRI
images are done using K-mean clustering.Inthelaststagethe
k-nearest neighbor classifier isusedtoclassifytheMRIimage
as cancerous or non cancerous. Also the twelve properties of
extracted featuresarecalculated.Thereforethisclassification
method becomes robust and effective .
In[7], Detection of brain tumorfromMRIimagesinvolves
different steps such as Magnetic Resonance image pre-
processing, segmentation of image feature extraction. This
paper describes about the methods that are used Histogram
Thresholding, K-means clustering, and Fuzzy C-Means
Support Vector Machine (SVM).Here includes several steps
such as pre-processing; high frequency components and
noise are removal RGB to gray conversion, global image
threshold; which converts intensity image into binaryimage,
erosion dilation of binary image to locate tumor position
exactly, detecting the stage of the tumor whether primary
Benign or last Malignant.
In[4], the authors proposed Brain tumor detection and
segmentation by using thresholding and watershed
algorithm. The main used was segmentation, which is done
usinga method based on threshold segmentation,watershed
segmentation and morphological operators. Threshold
segmentation is one of the simplest segmentation methods.
The input gray scale image is converted into a binary format.
The method is based on a threshold value which will convert
gray scale image into a binaryimage format.Themainlogicis
the selection of a threshold value. Some common methods
used under this segmentation include maximum entropy
method and k- means clustering method for segmentation.
Watershed Segmentation is one of the best methodstogroup
pixels of an image on the basis of their intensities. It is a good
segmentation technique for dividing an image to separate a
tumor from the image Watershed is a mathematical
morphological operating tool. After converting the image in
the binary format, some morphological operations are
applied on the converted binary image. The purpose of the
morphological operators is to separate the tumor part of the
image. Now only the tumor portion of the image is visible,
shown as white color. This portion has the highest intensity
than other regions of the image.
3. METHODOLOGY
Our proposed methodology is composed of a set of stages
starting from collecting MRI to classification as normal or
abnormal. The main steps are shown in Fig. 1.
Fig 1: Block diagram of the proposed method
MRI dataset
The dataset consists of T2-weighted, and axial, 256×256
pixel MRI. The dataset consists of 600, which 300 are of
abnormal MRI and 300 are of normal MRI. Fig. 2 shows

samples images for abnormal andnormal brain:(A) -Normal
brain (B) - Tumor brain.
Image Preprocessing
 Divides T2-Weighted MRI images into blocks with
size 60*80 using bounding box method. So we get
large number of dataset from each image.
 Then Instead of a single image here every blocks of
a image handled.
noise removal and smoothingwithpreservingedge
information.
Feature Extraction
Feature selection is the strategy of choosing a subset of
applicable features for structure powerful learning models
by evacuating most unimportant and excess features from
the data. The characteristic featuresoftheimagecanprovide
a helpful hand to the detection of tumor. Feature extraction
involves the process of collecting informationsuchas shape,
color, texture, gray level, contrast, etc . Texture analysis
gives an idea of the image . Moreover, statistical featurescan
be employed to have an overview of the given MRI image.
Features for Brain Tumor Detection:
A. Statistical Analysis
Statistical features the pixels or the gray level
values of the image gives spatial featuresandcanbe
used in bio-medical applications to study the image
characteristics. Statistical features include mean,
variance and standard deviation, which are briefly
stated as below:
 Mean: Mean of an image depends uponthe
homogeneity of brightness of the MRI
image. Mean will have high value if the
image is persistently bright. Mean is
defined as the sum of all pixel values
divided by total number of pixels. It gives
the average distribution of the intensity
values of an image.
 Variance: Variance characterizes
distribution of calculated gray levels. If
there is difference between gray level
values of means then the variance will be
increased. Variance gives a measure of
how each pixel differ from the mean value.
It is given as the average of square of the
difference between mean and individual
pixel.
 Standard deviation: Standard deviation
defines the difference in set of data values
from the mean. It can be expressed as the
square root of variance.
B. Gray-level co-occurrence matrix (GLCM)
The GLCM functions characterize the texture of an
image by calculating how often pairs of pixel with
speciﬁc values and ina speciﬁedspatial relationship
occur in an image. Gray Level Cooccurence Matrix
is one of the most widely used image analysis
application which can be employed to extract
textural features [8]. Feature extraction from
therapeutic images should be possible by following
two stages utilizing this strategy. The initial step
gives GLCM calculation. Texture highlights
dependent on GLCM is determined in second step
[8]. GLCM takes gray levels as input. Hence, the
given image is checked for a gray scale image. If it is
not, RGB image will be converted into gray scale
image to take gray levels as input. Textural findings
and analysis helps in assessment and diagnosis of
tumor and its stages. Using this matrix can measure
energy, contrast, homogeneity, Entropy and IDM.
 Energy: Energy is a portion which claims picture
similarity. It mirrors pixel pair replications. Gray
level is the fundamental unit of each image. Energy
is given as the sum of square of all gray levels or the
pixel values.
 Contrast: Contrast of an image returns the change
in the measure of intensity values between a pixel
and its neighbouring pixel. If there are many
variation in adjacent gray level alterations, it
implies high level of dissimilarity.
 Homogeneity: Homogeneity gives a measure of
value that measures the closeness o distribution of
elements in the GLCM to the GLCM diagonal.
 Entropy: Entropy is used to calculate dissimilarity
in MRI image or in the Region of interest (ROI).
Entropy is the statistical measure of randomness
that can be used to characterize texture of an
image.
 IDM: Inverse Difference Moment is the local
homogeneity. It is high when local grey level is
uniform.
C. Morphological Operation
The morphological operations are applied with
respect to shape of tumor. Basic operations are
Erosion and Dilation.

Erosion : Morphological erosion removes islands
and small objects so that only substantive objects
remain.
Dilation : Morphological dilation makes objects
more visible and fills in small holes in objects.
D. Edge
This is also essential feature required and sincethe
inherent nature of the brain tissues is such it not
easy to find the sharp boundaries of the tumor, this
nature and pixel arrangement of the edges is
considered for detection of the brain tumor. An
edge happens when there is an sudden and
unexpected intensity modification of the image.
Whenever it is detected an abrupt modification ora
change in the intensity of a certain image, the
associated pixel would be treated as an edge pixel.
Classification
Normal and abnormal MRIareclassifiedbytakingtheoutput
of feature extraction. Block matching technique with FFBP
used for training and testing. Using bounding box method
required less no of MRI images for training. Using stride get
huge dataset from each image. Also gives correct
classification and abnormal areas from images through
training with FFBP. Abnormal areas identified by bounding
box. Find the most dissimilar region mention it as Region of
Interest (ROI). Finally classifying both normal andabnormal
MRI images and gives area and location of tumor from
abnormal images.
Feed-ForwardBack-Propagation(FFBP)Neural Network
 Feed-Forward Neural Network.
In feed-forward neural network, the neurons are arranged
in layers and they have unidirectional connections between
them. They produce only one set of output values. They are
called as static network because in this the outputvalues are
produced only based on current input. The output values
does not depend on previous input values. They are also
called as memory less network. In feedback network, the
neurons have bidirectional connections between them.
Feedback or Recurrent networks produce a set of values
which depends on the previous input values. Feedback
network is also known as dynamic network because the
output values always depend on the previous input values.
 BPNN(Back-Propagation Neural Network)
BP neural network architecture with one hidden layer
operating on log sigmoid transfer function has been
employed for the classification of normal and abnormal
tumour. There is a full connectivity between the upper and
lower layers and no connections between neurons in each
layer. The weights on these connections encode the
knowledge of a network. The data enters at the input and
passes through the network, layer by layer, until it arrivesat
the output. The parameters of a network were adjusted by
training the network on a set of reference data, called
training set. The training of the network was performed
under back propagation of the error. The trained networks
were then be used to predict labels of the new data.
Algorithm stages for BPNN
1. Initialization of weights
2. Feed forward
3. Back propagation of Error
4. Updation of weights and biases
4. EXPERIMENTAL RESULTS
ABNORMAL BRAIN IMAGE:
At the point when gives abnormal image as input then get
right classification (Tumor Detected), also predict area and
size of tumor. It gives Grade of tumor in regards tothesize.If
tumor size less than 20, at that point that one in grade I, else
on the off chance that tumor size greater than 34, at that
point that one in grade III, else that kind of tumor in grade II
class.
Types of tumor:
Grade I includes : Benign
Grade II includes : Glioblastoma
Grade III includes : Malignant
Fig 2: Results of abnormal brain image as input
NORMAL BRAIN IMAGE:
At the point when gives normal image as input then get
correct classification (No Disease Effected).

Fig 3: Results of normal brain image as input
V. PERFORMANCE EVALUATION
Performance of proposed method is evaluated using
confusion matrix. The table shows the confusion matrix for
following data.
Total Images=600 Actual Abnormal Images=300 Actual
Normal Images =300
Table I : Confusion Matrix
Table II : Parameter Analysis
5. CONCLUSION
Medical images are amongst important data sources, since
these images are usually used by physicians to detect the
diseases that harm the human brain. Magnetic Resonance
Imaging (MRI) is a medical imaging technique used to
visualize the internal structure of the body and provide high
quality images. MRI contains useful and fine information
which is used to improve diagnosis accuracy.Thisautomatic
segmentation algorithm gives shape, size and location ofthe
tumor more accurately. It gives a less demanding approach
to analyze the tumor and encourages specialists to design
the careful methodology. Information of images can be
obtained by Statistical analysis, where the possibility of
tumor is highest by using mean, entropy and GLCM.
Bounding Box technique used to locating tumor areas. Easy
to extract features like size and location. Using boundingbox
method less number of MRI images requiredfortrainingand
also extracting features from each blocks.Whichhelpstofast
detection of abnormal areas. Block matching techniquewith
FFBP used to classification. Feed forward Back propagation
neural network is a multi-level error feed-back network that
reduces error rate. We have tested the proposed approach
using a benchmark dataset of MRI brain images. The
experimental results show that our approach achieved was
98 % classification accuracy achieved by ANN.
REFERENCES
[1] V. S. Takate P . S. Vikhe. ”Classification of MRI Brain
Images using KNN and k-means”, International Journal
on Advanced Computer Theory and Engineering
(IJACTE), 2014
[2] S.N.Deepa B.Aruna Devi. ”Artificial Neural Networks
design for Classification of Brain Tumor”, 2012
International Conference on Computer Communication
and Informatics , 2012
[3] N Muhammad Aqeel Aslam, Daxiang Cui. ” Brain
tumor detection and segmentation by using thresh-
olding and watershed algorithm”, ,Nano Biomed. Eng.,
2017
[4] Kanupriya Parveen, Amritpal Singh. ” BraTS: Brain
Tumor Segmentation SomeContempo-
raryApproaches”,IEEEInternationalConferenceonSignal
Processing and Integrated Networks (SPIN), pg. no. 98-
102, 2015
[5] Marjia Mandhir Kaur and Dr. Rinkesh Mittal.
”Survey of Intelligent Methods for Brain Tumor
Detection”, International Journal of Computer Science
Issues, Vol. 11, Issue 5, No 1, September 2014.
[6] K.V. Anitha and S. Murugavalli. ”Brain tumour
classification using twotier classifier with adaptive
segmentation technique”, IET Computer Vision, 2016,
vol.10(1), pp. 9-17.
[7] Hari Babu Nandpuru, Dr. S. S. Salankar, Prof. V. R.
Bora. MRI Brain Cancer Classifi-cation Using Support
Vector Machine, IEEE International Conference on
Electrical, ElectronicsandComputerScience,Pg.No.1-6,
1-2 March 2014.
[8] A.R. Kavitha, Dr. C.Chellamuthu, Ms. Kavin Rupa. An
Efficient ApproachforBrainTumourDetectionBasedon
Modified Region
[9] Takate, et al. and P. S. Vikhe , `` Classification of MRI
Brain Images using KNN and K-means”, IEEE,pp.-55-58,
2012.

[10] Deepa, et al. and B.A. Devi, `` Neural Networks
design for Classification of Brain Tumor”, The
international Conference on Computer Communication
and Informatics, Coimbatore, pp-568-573, 2012.
[11] M. Mohamed Fathima, D.Manimegalai, D.;
Thaiyalnayaki,S.,"Automatic detectionoftumorsubtype
in mammographs based on GLCM and DWT features
using SVM,"InformationCommunicationandEmbedded
Systems, International Conference on , vol., no.,
pp.809,813, 21-22 Feb. 2013
[12] N. M. Saad, S.A.R Abu-Bakar, S. Muda, N..M Mokji,
L. Salahuddin, "Brain lesion segmentation of
Diffusionweighted MRI using gray level co-occurrence
matrix," InInternational ConferenceonImagingSystems
and Techniques , pp. 284-289, 17-18 May 2011
[13] Heena Hooda, Om Prakash Verma, Tripti Singhal ,
“Brain Tumor Segmentation: A Performance Analysis
using K-Means, Fuzzy C-Means and Region Growing
Algorithm”, IEEE International ConferenceonAdvanced
Communication Control and Computing Technologies ,
pg. no. 1621-1626, 2014

IRJET- Brain Tumor Detection and Classification with Feed Forward Back Propagation Network

Related slideshows

More Related Content

IRJET- Brain Tumor Detection and Classification with Feed Forward Back Propagation Network