IRJET- Brain Tumor Detection using Deep Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 1460
Brain Tumor Detection using Deep Learning
Prof. B.S. Satpute1, Anand Kale2, Devashish Dhande3, Hitesh Kuber4, Saish Chore5
1 Professor, Department of Computer Engineering, DIT, Pimpri, Maharashtra, India
2,3,4,5Engineering Student, Department of Computer Engineering, DIT, Pimpri, Maharashtra, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Medical image processing is that the one
among the foremost demanding and promising field
nowadays. Tumor is a rapid uncontrolled growth of cell. The
tumor are often classified as benign, malignant and
premalignant. When a tumor is noticed as malignant then
the tumor results in cancer. Earlier stage of tumor is used to
be detected manually through observation of image by
doctors and it takes more time and sometimes gets
inaccurate results. Today different computer added tool is
employed in medical field. These tools provide a quick and
accurate result. Magnetic Resonance Images (MRI) is the
most widely used imaging technique for analyzing internal
structure of human body. The MRI is used even in diagnosis
of most severe disease of medical science like brain tumors.
The brain tumor detection process consist of image
processing techniques involves four stages. Image pre-
processing, image segmentation, feature extraction, and
finally classification. There are several existing of techniques
are available for brain tumor segmentation and
classification to detect the brain tumor. There are many
techniques available presents a study of existing techniques
for brain tumor detection and their advantages and
limitations. To overcome these limitations, propose a
Convolution Neural Network (CNN) based classifier. CNN
based classifier does the comparison between trained and
test data, from this to get the simplest result.
Keywords: Brain Tumor Detection, CNN, Image Pre-
processing
1. INTRODUCTION
Brain is that the management center within the physical
body. It is responsible to execute all activities throughout a
large number of connections and a huge number of
neurons. Brain tumor is one of the most serious diseases,
occurred due to an abnormal growth of cells in the brain,
affecting the functions of the nervous system. There are
different types of brain tumors which can be either
malignant or benign.
The early stage of tumor detection depends on the
physician’s knowledge and experience, making the
patients have a chance to recover his life and survival. An
automated classification system of brain tumors is an
effective tool for supporting the physicians to follow a
successful treatment option. Such system uses the images
captured by magnetic resonance (MR) imaging devices
which are widely used by the radiologists of brain
diagnosis.
Malignant brain tumors are usually in the form of blood
clots accompanied by fat surrounding it. To detect the
location and size of brain tumors required MRI images of
brain tumors. MRI images can help differentiate brain
tissue, brain tumors, edema, and spinal fluid supported
differences in color contrast in each tissue. The problem in
radiological remains analyzing the results of MRI brain
tumour manually in order that it takes an extended time to
seek out out the diagnostic from the doctor.
Image processing is a process of analyzing, manipulating a
picture so as to perform some operation to extract the
knowledge from it. According to world health
organization’s statistics, cancer is considered as the
second leading cause of human fatalities across the world,
being responsible for an estimated 9:6 million deaths in
this year. Among different form of cancers, brain tumor is
widely seen together of the deadliest cancers because of
its aggressive nature, heterogeneous characteristics
(types), and low relative survival rate (e.g., in US relative
survival rate following a diagnosis of a primary malignant
brain tumour is around 35%).
Medical imaging seeks for disclosure of internal structures
hidden by skin and bones and also to diagnose and treat
disease. And also, it establishes a database of normal
anatomy and physiology to form it possible to sense
abnormalities.
In today’s world, one of the reasons in the rise of mortality
among the people is brain tumor. Abnormal or
uncontrolled growth of cell developed inside the human
body is called brain tumor.
This group of tumor grows within the skull, due to which
normal brain activity is disturbed. Brain tumor is a serious
life frightening disease. So, which not detected in earlier
stage, can take away person’s life. Brain tumors are often
mainly three varieties called benign, malignant, pre-
malignant. The malignant tumor leads to cancer.
1.1 CNN
Convolutional neural network(CNN, or ConvNet) is type of
profound learning and most usually applied to dissecting
visual symbolism.CNNs utilize a variety of multilayer
perceptron's intended to require negligible pre-
processing.They are moreover insinuated as move
invariant or space invariant artificial neural
network(SIANN), supported their normal burdens
structure and translation invariance characteristics.

Convolutional networks were propelled by natural
procedures in that the availability design between
neurons takes after the association of the creature visual
territory . Individual cortical neurons answer stimuli only
during a restricted region of the field of vision referred to
as the receptive field. The receptive fields of various
neurons partially overlap such they cover the whole field
of vision .CNNs utilize moderately little pre-preparing
contrasted with other image classification algorithm. This
implies the system learns the channels that in customary
calculations were hand-built. This autonomy from earlier
information and human exertion in include configuration
might be a significant bit of leeway.They have applications
in image and video recognition, recommender systems,
image classification, medical image analysis, and tongue
processing.
A CNN consists of an input and an output layer, also as
multiple hidden layers. The hidden layers of a CNN
typically contains convolutional layers, pooling layers,
fully connected layers and normalization layers.
Fig1. Simple ConvNet.
There are four primary activities in the ConvNet appeared
in fig. above:
1. Convolution
2. Non-Linearity (ReLU)
3. Pooling or Sub Sampling
4. Classification (Fully Connected Layer)
The Convolution Step:
The main objective of Convolution is to gather features
from the input image. Convolution conserves the spatial
relationship within pixels by learning image features using
small squares of input data. Here every image can be
considered as a matrix of pixel values. Let’s consider a 5 x
5 image whose pixel values are only 0 and 1, the 5x5
matrix is a special case where pixel values are 0 and 1.
Also, consider another 3 x 3 matrix as kernel. Then, the
Convolution of the 5 x 5 image and the 3 x 3 matrix can be
computed as given below:
Fig2. The Convolution operation
The output matrix is called Convolved Feature. We slide
the orange matrix over our original image by 1 pixel
(called ‘stride’) and for every position, we compute
element wise multiplication and add the multiplication
outputs to get the final integer which forms a single
element of the output matrix.
As an example, consider the following input image: It is
evident from the animation above that different values of
the filter matrix will produce different Feature Maps for
the same input image. As an example, consider the
following input image:
Fig3. Sample Image
In the table below, we will see the consequences of
convolution of the above image with different filters. As
shown, we will perform operations like Edge Detection,
Sharpen and Blur just by changing the numeric values of
our filter matrix before the convolution operation– this
suggests that different filters can detect different features
from a picture , for instance edges, curves etc.

Fig4. Various Operation on Sample Image
Introducing Non-Linearity (ReLU):
An additional operation called ReLU has been used after
every Convolution operation. ReLU stands for Rectified
linear measure and may be a non-linear operation. Its
output is given by:
ReLU is an applied per pixel and replaces all pixels valued
negative in the feature map by zero. The objective of ReLU
is to present the non-linearity in our ConvNet, as majority
of real-world data we need our ConvNet to learn would be
non-linear (Convolution is a linear operation – element
wise matrix operation and addition, so we account for
non-linearity by presenting a non-linear function like
ReLU).
The Pooling Step:
Spatial Pooling reduces the dimensionality of each feature
map but keeps the important information. Spatial Pooling
are of numerous types like Average, Sum, Max etc.
In Max Pooling, we need to define a spatial neighborhood
and take the largest element from the corrected feature
map within that window. Instead of taking the largest
element we could also take the average or sum of all
elements in that window. In practice, Max Pooling has
been shown to figure better.
Shows an example of Max Pooling operation on a Rectified
Feature map (obtained after convolution + ReLU
operation) by employing a 2×2 window.
Fig5. Max Pooling
We slide our 2 x 2 window by 2 cells (also called ‘stride’)
and take the maximum value from each region. This
process reduces the dimensionality of our feature map.
1.2 SVM
Support Vector Machines (SVMs) square measure
supervised learning models with associated learning
algorithms that analyze knowledge used for classification
and statistical procedure. Given a gaggle of coaching
examples, each marked as belonging to a minimum of 1 or
the opposite of two categories, an SVM training algorithm
builds a model that assigns new examples to a minimum of
1 category or the opposite , making it a non-probabilistic
binary linear classifier (although methods like Platt scaling
exist to use SVM during a probabilistic classification
setting). An SVM model could even be a representation of
the examples as points in space, mapped so as that the
samples of the separate categories are divided by a
transparent gap that's as wide as possible. New examples
are then mapped into that very same space and predicted
to belong to a category supported the side of the gap on
which they fall.
A Support Vector Machine develops a hyperplane or set of
hyperplanes during a high-or unending dimensional space,
which can be utilized for characterization, relapse, or
different errands like exceptions detection.[3] Intuitively,
a fair partition is accomplished by the hyperplane that has
the first significant separation to the nearest preparing
information purpose of any class (supposed useful edge),
since for the most part the bigger the edge, the lower the
speculation blunder of the classifier. A Support Vector
Machine (SVM) could even be a discriminative classifier
formally defined by a separating hyperplane. As it were,
given named preparing information (managed learning),

the calculation yields an ideal hyperplane which sorts new
models. In two dimensional space this hyperplane could
even be a line dividing a plane in two parts where in each
class lay in either side.
Fig6. Hyperplanes
2. Literature Survey
[1] Abdu Gumaei, Mohammad Mehedi Hassan
Cerebrum malignant growth characterization could
likewise be a significant advance that relies upon the
doctor's information and information. an automatic tumor
arrangement is extremely essential to support radiologists
and physicians to spot brain tumors. However, the
accuracy of current systems got to be improved for
suitable treatments. during this paper, we propose a
crossover highlight extraction strategy with regularized
outrageous learning machine for building up an exact
cerebrum tumor arrangement approach. The methodology
begins by separating the highlights from brain pictures
utilizing the mixture include extraction technique; at that
point, registering the covariance network of those
highlights to extend them into a replacement significant
set of features using principle component analysis (PCA).
Finally, a regularized extreme learning machine (RELM) is
employed for classifying the sort of brain tumour . to
guage and compare the proposed approach, a gaggle of
experiments is conducted on a replacement public dataset
of brain images. Experimental results proved that the
approach is simpler compared to the existing state-of-the-
art approaches, and thus the performance in terms of
classification accuracy improved from 91.51% to 94.233%
for the experiment of random holdout technique.
[2] Annisa Wulandari, Riyanto Sigit
Brain tumor is one among disease type that attacks the
brain within the sort of clots. there's how to ascertain
brain tumour intimately requires by an MRI image. there's
difficulty in distinguishing brain tumour tissue from
normal tissue due to the similar color. brain tumour must
be analyzed accurately. the answer for analyze brain
tumour is doing segmentation. brain tumour segmentation
is completed to separate brain tumour tissue from other
tissues like fat, enema, normal brain tissue and spinal fluid
to beat this difficulty, The MRI image must be maintained
at the sting of the image first with the median filtering.
Then the tumor segmentation process requires
thresholding method which is then iterated to require the
most important area. The brain segmentation is completed
by giving a mark on the world of the brain and areas
outside the brain using watershed method then clearing
skull with cropping method. during this study, 14 brain
tumour MRI images are used. The segmentation results
are compared brain tumors area and brain tissues area.
this technique obtained the calculation of tumor area has a
mean error of 10%.
[3] Mircea Gurbin
The brain is one among the foremost complex organs
within the physical body that works with billions of cells. A
cerebral tumor occurs when there's an uncontrolled
division of cells that form an abnormal group of cells
around or within the brain. This cell group can affect the
traditional functioning of brain activity and may destroy
healthy cells. Brain tumors are named considerate or
second rate (grade 1 and 2) and dangerous tumors or high
(grade 3 and 4). The proposed strategy expects to separate
between ordinary cerebrum and tumor brain (considerate
or insult). The study of some sorts of brain tumors like
metastatic bronchogenic carcinoma tumors, glioblastoma
and sarcoma are performed using brain resonance
imaging (MRI). The detection and classification of MRI
brain tumors are implemented using different wavelet
transforms and support vector machines. Accurate and
automatic classification of MRI brain images is extremely
important for medical analysis and interpretation.
[4] Parnian Afshar, Konstantinos N. Plataniotis
According to official statistics, cancer is taken into account
because the second leading explanation for human
fatalities. Among varying kinds of malignant growth, brain
tumor is seen together of the deadliest structures on
account of its forceful nature, heterogeneous qualities, and
low relative endurance rate. Determining the sort of brain
tumour has significant impact on the treatment choice and
patient’s survival. Human-focused finding is normally
blunder inclined and questionable prompting an ongoing
flood important to automatize this procedure utilizing
convolutional neural systems (CNNs). CNNs, be that as it
may, neglect to totally use spatial relations, which is
particularly destructive for tumor grouping, in light of the
fact that the connection between the tumor and its
encompassing tissue might be a basic marker of the
tumor's sort. In our recent work, we've incorporated
newly developed CapsNets to beat this shortcoming.
CapsNets are, however, sensitive to the miscellaneous
image background. The paper addresses this gap. the most
contribution is to equip CapsNet with access to the tumor
surrounding tissues, without distracting it from the most
target. An adjusted CapsNet architecture is, in this manner,
proposed for brain tumor characterization, which takes
the tumor coarse limits as additional contributions inside
its pipeline to expand the CapsNet's core interest. The
proposed approach discernibly outflanks its partners.

[5] T. M. Shahriar Sazzad, Misbah Ul Hoque
A tumor cell may be a sort of cell that develops out of
control of the standard forces and standardizes growth.
brain tumour is one among the main reasons for fatality
per annum . Around 50% of brain tumour diagnosed
patient die with primary brain tumors annually within the
us . Electronic modalities are wont to diagnose brain
tumors. Among all electronic modalities, resonance
Imaging (MRI) is one among the foremost used and
popular for brain tumour diagnosis. during this research
study, an automatic approach has been proposed where
MRI gray-scale images were incorporated for brain
tumour detection. This study proposed an automatic
approach that has enhancement at the initial stage to
attenuate gray-scale color variations. Filter operation was
wont to remove unwanted noises the maximum amount as
possible to help better segmentation. As this study test
grayscale images therefore; threshold based OTSU
segmentation was used rather than color segmentation.
Finally, pathology experts provided feature information
was wont to identify the region of interests (brain tumor
region). The experimental results showed that the
proposed approach was ready to perform better results
compared to existing available approaches in terms of
accuracy while maintaining the pathology experts’
acceptable accuracy rate.
[6] Chuanlu Lin, Yi Wang, Tianfu Wang, Dong Ni
We present a typical system for the synchronous division
and recuperation of obsessive magnetic resonance (MR)
brain pictures, where low-rank and sparse decomposition
(LSD) plans have been generally utilized. Conventional
LSD methods often produce recovered images with
distorted pathological regions, due to the lack of
constraint between low-rank and sparse components. To
address this issue, we propose a transformed low-rank
and structured sparse decomposition (TLS2D) method,
which is robust for extracting pathological regions. Also,
the all around recuperated pictures can be acquired
utilizing both organized inadequate and figured picture
saliency as the versatile sparsity requirement.
Experimental results on MR brain tumor images
demonstrate that our TLS2D can effectively provide
satisfactory performance on both image recovery and
tumor segmentation.
[7] Sérgio Pereira, Adriano Pinto, Victor Alves
Among brain tumors, gliomas are the foremost common
and aggressive, leading to a very short anticipation in their
highest grade. Thus, treatment planning could also be a
key stage to reinforce the quality of lifetime of oncological
patients. Magnetic resonance imaging (MRI) could
likewise be a broadly utilized imaging method to survey
these tumors, however the gigantic measure of
information delivered by MRI forestalls manual division
during an inexpensive time, limiting the use of precise
quantitative measurements within the clinical practice.
Along these lines, programmed and dependable division
strategies are required; be that as it may, the immense
spatial and basic fluctuation among brain tumors make
automatic segmentation a difficult issue. during this paper,
we propose an automatic segmentation method supported
Convolutional Neural Networks (CNN), exploring small
3x3 kernels. the utilization of little bits permits structuring
a more profound design, other than having a constructive
outcome against overfitting, given the less number of
loads inside the system. We additionally researched the
utilization of force standardization as a pre-handling step,
which however not normal in CNN-based division
strategies, demonstrated related to information expansion
to be exceptionally viable for mind tumor division in MRI
pictures. Our proposition was approved inside the mind
tumor Segmentation Challenge 2013 database (BRATS
2013), getting all the while the principal position for the
entire , center, and improving districts in Dice Similarity
Coefficient metric (0.88, 0.83, 0.77) for the Challenge
informational collection. Additionally, it acquired the
general first situation by the online assessment stage. We
additionally took an interest inside the on location BRATS
2015 Challenge utilizing a similar model, acquiring the
subsequent spot, with Dice Similarity Coefficient metric of
0.78, 0.65, and 0.75 for the entire, center, and upgrading
areas, separately.
3. Proposed System
As per literature survey, it had been found that automated
brain tumour detection is extremely necessary as high
accuracy is required when human life is involved.
Automated detection of tumor in MR images involves
feature extraction and classification using machine
learning algorithm. Our approach consists of three steps:
(A) Brain image pre-processing, (B) Brain feature
extraction, and (C) brain tumour classification. The input
of the approach is that the brain images and therefore the
output are the respective sort of the brain tumour. The
small print of the steps of our proposed approach are
described within the subsections below. during this paper,
a system to automatically detect tumor in MR images is
proposed as shown in figure 3.

Fig7. Proposed System
A. Image Pre-processing
Pre-preparing might be a typical name for activities with
pictures at absolute bottom degree of deliberation both
info and yield are intensity images.
The purpose of pre-handling with is an improvement of
the picture data that smothers unfortunate curves or
redesigns some picture features huge for extra handling.
The reason for picture preparing is separated into 5
gatherings.
They are:
1. Perception - Observe the articles that are not obvious.
2. Picture honing and rebuilding - To make a superior
picture.
3. Picture recovery - Seek for the picture of intrigue.
4. Estimation of example – Measures different articles in a
picture.
5. Picture Recognition – Distinguish the items in a picture.
a. Gray Scale
Grayscale picture otherwise called highly contrasting
picture is the one wherein every pixel of the picture
conveys power data. Dim scale picture has just two hues:
Black and white. The changed over grayscale picture may
lose contrasts, sharpness, shadow, and structure of the
shading picture. The luminance of a pixel estimation of a
grayscale picture ranges from 0 to 255.
Fig8.Gray Scale Image
b. Smoothing
The low-pass filters for the most part utilize moving
window administrator which influences each pixel of the
picture in turn, changing its incentive by some capacity of
a local region (window) of pixels. The administrator
moves over the picture to influence all the pixels in the
picture. The operator moves over the image to affect all
the pixels in the image.
Fig 9. Smoothing
c. Edge Detection
Watchful edge recognition is a used to remove valuable
basic data from various articles and lessen the measure of
information to be handled. The general criteria for edge
recognition incorporate: -
1. Detection of edge with low blunder rate, which
implies that the discovery ought to precisely get however
many edges appeared in the picture as could reasonably
be expected.
2. The edge point identified from the administrator
ought to precisely confine on the focal point of the edge.
3. A given edge in the picture should just be checked
once, and where conceivable, picture clamor ought not
make bogus edges.
Fig10.Edge Detection
4. CONCLUSION
In summary, we propose a CNN-based method for
segmentation of brain tumors in MRI images. There are
several existing of techniques are available for brain
tumor segmentation and classification to detect the brain

tumor. There are many techniques available presents a
study of existing techniques for brain tumor detection and
their advantages and limitations. To overcome these
limitations, propose a Convolution Neural Network (CNN)
based classifier. CNN based classifier used to compare
the trained and test data, from this get the best result.
REFERENCES
[1] D. Kornack and P. Rakic, “Cell Proliferation without
Neurogenesis in Adult Primate Neocortex,” Science, vol. 294,
Dec. 2001, pp. 2127-2130, doi:10.1126/science.1065467.
[2] M. Young, The Technical Writer’s Handbook. Mill Valley,
CA: University Science, 1989.
[3] R. Nicole, “Title of paper with only first word capitalized,”
J. Name Stand. Abbrev., in press.
[4] K. Elissa, “Title of paper if known,” unpublished.
[5] Zeynettin Akkus, Alfiia Galimzianova, Assaf Hoogi , Daniel
L. Rubin and Bradley J. Erickson, “Deep Learning for Brain
MRI Segmentation: State of the Art and Future Directions” J
Digit Imaging DOI 10.1007/s10278-017- 9983-4, 2017.
[6] Anupurba Nandi, “Detection of human brain tumour using
MRI image segmentation and morphological operators” IEEE
International Conference on Computer Graphics, Vision and
Information Security (CGVIS), 2015.
[7] Swapnil R. Telrandhe, Amit Pimpalkar and Ankita Kendhe,
“Detection of Brain Tumor from MRI images by using
Segmentation &SVM” World Conference on Futuristic Trends
in Research and Innovation for Social Welfare (WCFTR’16),
2016.

IRJET- Brain Tumor Detection using Deep Learning

Recommended

More Related Content

What's hot (20)

Similar to IRJET- Brain Tumor Detection using Deep Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Brain Tumor Detection using Deep Learning