Brain Tumour Detection Using CNN
Sri lekha Jagannadham
1*
, K. Lakshmi Nadh
2@
, M. Sireesha
3#
Department of CSE, Narasaraopeta Engineering College, Narasaraopet, Andhra Pradesh
1
srilekhasunny@gmail.com,
2
lakshminadh37@gmail.com,
3
sireeshamoturi@gmail.com
Abstract - The perilous disease in world nowadays is brain tumor.
Tumor will occur when the healthy tissues are damaged and affects the
brain. Tumor is the unlimited growth of bizarre cells in brain. Hence,
death will be caused when there is a rapid growth of tumor cells.
Detecting the abnormal tissues from normal brain tissues is the pivotal
role for brain tumor detection system. M ain thing is that the concepts of
medical image processing, M R images, these are mainly utilized by
domain of brain tumor analysis. Detecting and diagonising the brain
tumor in early stage is the significant task which can save a patient from
worse effects. In our work, the input is MRIs (Magnetic Resonance
Images) and from the input this research work attempts to extract tumor
cells. Some of the pre-processing techniques of deep learning are used
for removal of noise from the input images, to make the results more
accurate, this research work will apply augmentation techniques to
increase the training set and apply different Convolution Neural
Network [CNN] techniques to grab out the best details from the image.
In order to enhance the details from the image, the outer portion of the
image cut out from all 4 sides.
Keywords: Tumor, bizarre cells, Magnetic Resonance Image.
1 INTRODUCTION
Brain tumor is one of the rigorous diseases in the
medical science domain. Group of cells are abnormal and they
are formed from uncontrolled division of cells, which is also
called as tumor and these are formed and spread to spinal cord
and different cells of brain [14][26]. Main key concern of a
radiologist is the effective and efficient analysis of premature
phase of tumor growth. The tumors are assorted into two, benign
and cancerous. If tumor is not properly diagnosed and treated
there is a chance of causing death [2]. The tumor is an unusual
growth of tissues and uncontrolled cells and its rise [27]. Based
on the stereotactic biopsy test histological grading is the gold
standard and convention for detecting the grade of brain tumor.
In biopsy procedure neurosurgeon will drill a small hole into
skull where the tissue is collected. Biopsy test has many risk
factors, the factors are bleeding from tumor and brain causing
infection, seizures, severe migrane, stroke, coma and even death.
Stereotactic biopsy’s main concern is that it is not 100% accurate
and also result in serious diagnostic error followed by wrong
clinical management of the disease. Using MRI, this research
work obtains the brain images, and it can also recognise the noise
and any changes using these MRI’s during acquisition [16].
MRI (Magnetic Resonance Imaging), it is an imaging
technique and extensively used for diagnose the patient and
treatment of tumors in clinical practice [18][20][22]. The image
contains non-invasive soft tissue and this is used for diagnosis of
tumors within the brain [1]. MR images are taken in three
directions and they are sagittal, axial and coronal. MRI contain a
noise caused by operator performance which can lead to serious
inaccuracies classification [21]. The users can learn the pattern of
brain tumor using deep learning techniques because manual
segmentation is time consuming and also being susceptible to
human errors or mistakes. Computer detection systems are
challenging aspects in every field and still there is an open
problem because of difference in shapes, areas, and sizes of
tumor [3].
Techniques which are mentioned above will give
segmented MRI without restricting the region of tumor; some of
the techniques [24] are able to detect the single tumor but no
technique will address the localization and detection of very
small tumors. MRIs are majorly used to detect and visualize the
details in internal structure of body [25]. In our paper, we
proposed an automatic technique that detects the multiple and
also very small tumors. The contribution and novelty of proposed
system is that, it can detect localize and multiple tumors and
small tumors which are not recognized in manual segmentation
and the proposed model uses the same MRI image which is used
to identify the tumor manually no need to change of that images.
The segmentation should separate the active tumorous tissue
from the necrotic tissue, and also the edema (swelling near the
tumor) should be identified. This process is done by identifying
the areas which are abnormal when compared to normal tissue
[6].
When applying the classic segmentation methods and limitations
such as inhomogeneous intensity, complex physiological
structure and blurred tissues boundaries in brain MR images
usually lead to unsatisfactory results [17]. So, we are using 2
different Activation functions to build the model accurately and
different layers for creating CNN. The Activation functions are
Relu and Sigmoid. To get the better results we are using 200
epochs to train the model. During each epoch training, it’ll save
the model file in the directory. We can choose the best model to
predict the results for the test data.
2 LITERATURE SURVEY
Many researches has shown that, they worked on image
processing and soft computing has a review and analysis on
detection and augmentation[28] of brain tumour techniques.
There are many other researchers working on the brain tumour
detection techniques[29] because it is one of the most searching
and challenging task. At present segmentation based on Neural
networks became more prominent and increasing each and
everyday
The process of segmenting based on the Morphological
operations and SFCM algorithm which also makes the
computation time more. Our model(proposed) detects the tumour
with almost 92% accuracy. This process of segmenting with
MMO is established by Devkota [7]. Based on the segmentation
[26] of histogram technique on the detection of edge approach is
processed by Yanatao[8]. Dina [11] has introduced PNN [24]
model, which is related to Vector Quantization. Some other
authors has also implemented the PNN on segmenting technique
along with PCA, which can reduces the dimensionality and helps
in improving the feature extraction.
Proceedings of the Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC)
IEEE Xplore Part Number: CFP21OSV-ART; ISBN: 978-1-6654-2642-8
978-1-6654-2642-8/21/$31.00 ©2021 IEEE 734
2021 Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) | 978-1-6654-2642-8/21/$31.00 ©2021 IEEE | DOI: 10.1109/I-SMAC52330.2021.9640875
Authorized licensed use limited to: Indian Institute of Space Science And Technology. Downloaded on March 25,2022 at 04:26:36 UTC from IEEE Xplore. Restrictions apply.