International Journal of Advances in Engineering and Management (IJAEM)

Volume 4, Issue 4 Apr 2022, pp: 41-45 www.ijaem.net ISSN: 2395-5252

Brain Tumor Detection using Deep

Learning
Sadaf Aafreen1 , Iqura Zarreen2 ,Aftab Ahemad3 ,
1,2,3
Prof. Abdul Razzaque4
Students, Anjuman College of Engineering and Technology, Nagpur, Maharashtra
4
Associate Professor, CSE Department, Anjuman College of Engineering and Technology, Nagpur,
Maharashtra

---------------------------------------------------------------------------------------------------------------------------------------
Submitted: 01-04-2022 Revised: 04-04-2022 Accepted: 07-04-2022
----------------------------------------------------------------------------------------------------------------
ABSTRACT: This research paper is to detect Tumors can have different origins and based on the
Brain Tumor using Deep Learning. A brain tumor cells or the origin obtained from different types of
is a disease caused due to the growth of abnormal tumors. Symptoms of a brain tumor include
cells in the brain. There are two main categories of headache which can be acute and persistent,
brain tumor, they are non-cancerous (benign) brain muscular disorders, cognitive disorders, etc.
tumor and cancerous (malignant) brain tumor. With The presence of tumor is noticed by the
the growth of Artificial Intelligence, Deep learning Magnetic Resonance Imaging [MRI] scanning. The
models are used to diagnose the brain tumor by MRI scanning should be diagnosed by the
taking the images of Magnetic Resonance Imaging physician and later based on the results; the
(MRI). The research work carried out uses Deep treatments shall be started. In recent times,
learning models like convolutional neural network Computer-aided diagnosis of diseases is gaining
(CNN) model to detect the tumor region in the interest and is helping doctors take swift decisions.
scanned brain image. Convolutional Neural One such approach is using Convolutional Neural
Network (CNN) was employed for the task of Networks (CNN) to learn the spatial and temporal
classification.we have concentrate on deep learning features from the given dataset which are necessary
through brain tumor detection using normal brain to identify the disease. Convolutional Neural
image or abnormal by using deep learning. The Network (CNN) was employed for the task of
brain is largest and most complex organ in human classification. we have concentrate on deep
body that works with billions of cells. The learning through brain tumor detection using
convolutional neural network algorithm is used to normal brain image or abnormal by using deep
detecting the brain tumor. There are many existing learning. The convolutional neural network
techniques are available for brain tumor algorithm is used to detecting the brain tumor.
segmentation and classification to detect the brain The proposed technique has used CNN to
tumor. There are many techniques available identify and categorize the tumor from brain
presents a study of existing techniques for brain images of the brain. The main difference between
tumor detection and their advantages and the main channel of the neural network with the
limitations. To overcome these limitations, we used normal neural network is that it is able to
Convolution Neural Network (CNN) based automatically and locally extract the feature from
classifier. CNN based classifier used to compare each image. These types of networks consist of
the trained data and test data, from this data get the neurons with weights and biases that can be
best result. learned. A Convolutional Neural Network is a
KEYWORDS: CNN, MR Images, transfer special type of neural network which specializes in
learning, Deep Learning, Brain tumor etc. handling image datasets. The very fundamental
principle of this neural network is performing a
I. INTRODUCTION convolution operation between the kernel and the
A Brain Tumor is a mass of tissue in image to extract the features.
which the cells multiply uncontrollably. It arises Here we are detecting Brain tumor from
from different cells - both in the brain and outside. MR Images. There is an image dataset consisting

DOI: 10.35629/5252-04044145 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 41
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 4 Apr 2022, pp: 41-45 www.ijaem.net ISSN: 2395-5252

some MR Images of healthy brain and some MR Images.

Images containing brain tumor. Following are some

Fig: Magnetic Resonance Image (MRI) consisting Brain tumor[1]

DEEP LEARNING :
Deep learning[2] is a type of machine CONCEPT OF PERCEPTRON :
learning and artificial intelligence (AI) that imitates The Perceptron[3] is a Deep Learning
the way humans gain certain types of knowledge. Algorithm that Allow to built a binary
Deep learning is an artificial intelligence (AI) classification and a fun fact of perceptron That it is
function that initiates the workings of the human inspired by actual brain neuron. Just like a neuron
brain in processing data and creating patterns for in a body perceptron accept the multiple input in to
use in decision making. the form of matrix and than based on a probability
In ML Algorithms everything is flatten distribution value we can conclude which class to
and in single dimension array but whereas in deep the belong. A Perceptron is an algorithm used for
learning we use some think called as tenser and supervised learning of binary classifiers. Binary
tenser has basically small matrices inside a big classifiers decide whether an input, usually
matrix, so it can be consider as matrix nested at represented by a series of vectors belongs to a
inside a matrix specific class in short, a perceptron is a single-layer
Deep learning is an important element of neural network.
data science, which includes statistics and Convolutional Neural Network
predictive modeling. It is extremely beneficial to (ConNet/DNN) is a Deep Learning algorithm
data scientists who are tasked with collecting, which can take in an input image, assign
analyzing and interpreting large amounts of data; importance (learnable weights and biases) to
deep learning makes this process faster and easier. various aspects/objects in the image and be able to
When there is lack of domain understanding for differentiate one from the other.
feature introspection, Deep Learning techniques A Convolutional Neural Network is a
outshines others as you have to worry less about special type of neural network which specializes in
feature engineering. Deep Learning really shines handling image datasets. A convolutional layer is
when it comes to complex problems such as image often coupled with a pooling layer and we can
classification natural language processing, and connect multiple such convolutional layer-pooling
speech recognition. layer pairs. In the end, we can have few Dense
At its simplest, deep learning can be layers and dropout layers for the final learning
thought of as a way to automate predictive process.
analytics. While traditional machine learning
algorithms are linear, deep learning algorithms are
stacked in a hierarchy of increasing complexity and TRANSFER LEARNING CONCEPT:
abstraction. Transfer learning[4] for machine learning
DOI: 10.35629/5252-04044145 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 42
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 4 Apr 2022, pp: 41-45 www.ijaem.net ISSN: 2395-5252

is when existing models are reused to solve a new augmentation concept is introduced. The proposed
challenge or problem. The knowledge developed technique first converts each input MR modality to
from previous training is recycled to help perform a slices, and intensities are preprocessed using a
new task. The new task will be related in some way statistical normalization approach.K-means
to the previously trained task, which could be to clustering approach is implemented to segment
categorise objects in a specific file type. The brain tumors to focus ROI for precise feature
original trained model usually requires a high level extraction. Finally, to classify brain tumors into
of generalisation to adapt to the new unseen data. their two general classes (benign/malignant); a
Transfer learning means that training finetuned VGG-19 CNN model is trained perfectly
won’t need to be restarted from scratch for every using synthetic data augmentation techniques. The
new task. Training new machine learning models proposed CNN based method is evaluated by
can be resource-intensive, so transfer learning conducting rigorous experiments on 2015 data set.
saves both resources and time. The accurate Thus, the results show that the proposed technique
labelling of large datasets also takes a huge amount could assist the radiologist and medical expertsin
of time. The majority of dataencountered by detecting brain tumors and classifying them into
organisations can often be unlabelled, especially their respectiveclasses (benign/malignant). The
with the extensive datasets required to train a proposed computer analysis'sefficiency and
machine learning algorithm. With transfer learning, accuracy to design (CAD) system are
a model can be trained on an available labelled comparedwith recent existing methods and the
dataset, then be applied to a similar task that may results exhibited that theproposed technique
involve unlabelled data. exhibited better accuracy.
ChirodipLodh Choudhury and
II. LITERATURE REVIEW ChandrakantaMahanty[9], In this research paper,
Sneha Grampurohit[5]proposed work in They proposed a new system based on CNN, which
which Deep neural networks such as CNN and discriminates between the Brain MRI images to
VGG-16 are investigated on MRI images of Brain. mark them as tumorous or not. The model is having
Both the models have given an effectiveresult, CNN with 3 layers and requires very few steps of
However VGG-16 takes a greater computational pre-processing to produce the results in 35 epochs.
time and memory but has given satisfactory results The purpose of the research is to highlight the
compared to CNN. Due to the availability of huge importance of diagnostic machine learning
data being produced and stored by the medical applications and predictive treatment. To detect
sector, Deep learning will play an important role in brain tumour with neutrosophical principles in the
data analysis in the upcoming days. future using the Convolutional Neural Network.
Sobhangi Sarkar[6],The paper discusses Janki Naik [10]discussed“Tumor
the method for detecting abnormalities in the brain Detection and Classification using Decision Tree in
MRI images. Sarkar discussed and implemented a Brain MRI” is used to get accurate and efficient
deep learning architecture by leveraging result. Using Decision tree classification technique
convolutional neural networks for the classification tumor has been found as well as classified in
of the different types of brain tumor from MR Normal or Abnormal class. Here we used two
images. The model developed in the study plotted algorithms, They developed brain tumor
an accuracy of 91% and an overall precision and classification system is expected to provide
recall of 91% and 88% respectively. valuable diagnosis techniques for the physicians.
Dr. ChintaSomeswararao[7], This paper Khurram Shahzad and Imran
was a combination of CNN model classification Siddique[11], proposed an easy, fully automatic
problem for predicting whether the subject has and efficient algorithm for extraction of brain
brain tumor or not & Computer Vision problem for tumor has been introduced. Morphological
automate the process of brain cropping from MRI operation like erosion and dilation along with
scans. The final accuracy is much higher than 50% morphological gradient and threshold are used.
baseline (random guess). However, it could be Morphological gradient is used for calculating
increased by larger number of train images or threshold. Threshold is used to binarize the image
through model hyper parameters tuning. which results an image having tumor and some
Amjad Rehman Khan[8],This article has noise with it. Erosion is used for thinning the image
exhibited a comprehensive brain tumor as it shrinks the image and helps to reduce noise or
segmentation system and classification using unwanted small objects. Dilation is being used after
VGG19 CNN model on MRI data. To enhance the erosion so that to get removed tumor portion back
accuracy of the classifier synthetic data, which was being removed by erosion.

DOI: 10.35629/5252-04044145 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 43
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 4 Apr 2022, pp: 41-45 www.ijaem.net ISSN: 2395-5252

Aryan Sagar Methil,[12] This paper and gradients. Finally, Appliedpost-processing to

presents a novel method involving image improve our results.
processing techniques for image manipulation GajendraRaut[16], In this paper the time-
which would aid our CNN model to classify tumor consuming process of brain tumor detection is thus
and non-tumor images better. Image Processing simplified by automation. After detecting the tumor
techniques helped to solve the illumination issues with convolutional neural networks segmentation
and brought the tumor into focus. Data techniques like autoencoders and K-means are
augmentation was used to reduce the chances of applied over the tumorous image to locate the
overfitting, as it artificially expands the size of a region of tumor in the image.when segmented the
training dataset, thus bringing out an improvement tumor image directly with K-means it sometimes
in the performance and the ability of the model to produces a noisy poor segmented image. Hence for
generalize. There are limitations to this work as segmentation they combined Autoencoders with K-
there are small chances that the image pre- means which produced more precise and clear
processing applied can damage the information segmented images with less noise. Thus, an
which makes a tumor image appear non-tumor in efficient model for detection and segmentation of
the eye of the CNN model.For future brain tumor is build which saves human efforts and
improvements, we can use ensemble techniques time.
and combine the performance of different models G.Hemanth[17] proposed method employs
for better performance. a mean field term within the standard CNN
Swathi K and Kishore Balasubramanian objective function. The technique isdeveloped and
[13]propsed the automated segmentation applied in MATLAB environment byutilizing the
techniques provide a wide range of applications image processing tool. Datasets areassembledfrom
like image guided surgery, volume visualization of the UCI datasets. A comparison is portrayed among
regions of interest, medical diagnosis and serves an all the features and the entire result being depicted
aid to detect other neurological diseases. Though in thefigures. The accuracy is computed which is
automated, it requires verification of results from a then compared with rest of the state-of-arts
doctor (to be certified by a competent medical methods. Efficiency andtraining accuracy of the
professional before starting treatment). It is also proposed brain tumorclassification approach is
seen that accuracy obtained by individual methods computed.
on an average is not convincing. Hybrid algorithms
may reduce time complexity further, give accurate III. CONCLUSION
area of tumor occupied and aim in improving the In this brain tumor detection we have
accuracy, sensitivity and specificity studied about feature based existing work. In
Masoumeh Siar and Mohammad feature based we have study about image
Teshnehlab[14],In this paper they used the processing techniques likes image pre-processing,
combination of feature extraction algorithm and the image segmentation, features extraction,
CNN for tumor detection from brain images is classification. And also study about deep learning
presented. The CNN is capable of detecting a techniques CNN and VGG16.In this system we
tumor. The CNN is very useful for selecting an have detect the tumor is present or not if the tumor
auto-feature in medical images. Images collected at is present then model return’s yes otherwise it
the centers were labeled by clinicians, then, tumor return no. and we have compared CNN with the
screenings were categorized into two normal and VGG 16 Model. The result of comparison VGG 16
patient classes. is more accurate than CNN. However, not every
IlyasseAboussaleh[15] proposed an task is said to be perfect in this development field
approach based on CNN architecture in order to even more improvement may be possible in this
predict and segment simultaneously a cerebral application. We have learned so many things and
tumor. In this process, an MRI image was pre- gained a lot of knowledge about development field.
processed and augmented using normalization and
data augmentation techniques. The MRI image was IV. ADVANAGES
classified into a tumor or not tumor brain image by It has accuracy near about 95% using transfer
a CNN model with two neurons in the output layer; learning concept. Our system will be fast and
in this task,He used the ground truth to label the accurate fully automatic method for brain tumor
images as tumor or not tumor images. The detection.
segmentation was applied on the images that Transfer learning has the benefit of decreasing the
contained the tumor, using the features extracted training time for a neural network model and can
from the last convolution layer of CNN architecture result in lower generalization error.

DOI: 10.35629/5252-04044145 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 44
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 4 Apr 2022, pp: 41-45 www.ijaem.net ISSN: 2395-5252

Useful Learned Features: The models have [6]. Sobhangi Sarkar “A CNN based Approach
learned how to detect generic features from for the Detection of Brain Tumor Using
photographs, given that they were trained on more MRI Scans” Research Gate [June 2020]
than 1,000,000 images for 1,000 categories. [7]. Dr. ChintaSomeswararao“Brain Tumor
State-of-the-Art Performance: The models Detection Model from MR Images using
achieved state of the art performance and remain Convolutional Neural Network”
effective on the specific image recognition task for IEEMay[June 2020]
which they were developed. [8]. Amjad Rehman Khan “Brain tumor
Easily Accessible: The model weights are segmentation using K-means clustering and
provided as free downloadable files and many deep learning with synthetic data
libraries provide convenient APIs to download and augmentation for classification” JMET
use the models directly. [February 2020]
[9]. ChirodipLodh Choudhury and
V. APPLICATION ChandrakantaMahanty “Brain Tumor
 The main aim of the applications is tumor Detection and Classification Using
identification. Convolutional Neural Network and Deep
 With the help of our project we can detect the Neural Network” IEE [July 2020]
brain tumor. This system Can assist the [10]. Janki Naik “Tumor Detection and
physicians to make early decision so that the Classification using Decision Tree in Brain
treatment are carried out at an earlier stage. MRI” IJERD [2013]
 The reason behind the development of this [11]. Khurram Shahzad and Imran Siddique
application is to provide proper treatment as “Efficient Brain Tumor Detection Using
soon as possible and protect the human life Image Processing Techniques “International
which is in danger. Journal of Scientific & Engineering
 This application is helpful to doctors as well as Research [December 2019]
patient. [12]. Aryan Sagar Methil “Brain Tumor Detection
 The manual identification is not so fast, more using Deep Learning and Image Processing
accurate and efficient for user. To overcome “IEE [June 2021]
those problem this application is design. [13]. Swathi K, Kishore Balasubramanian
“Preliminary investigations on automatic
 It is user friendly application.
segmentation methods for detection and
volume calculation of brain tumor from MR
REFERENCES images “Biomedical Research [February
[1]. Navoneel Chakrabarty “Brain MRI images 2016]
dataset” [April 2019]
[14]. Masoumeh Siar “Brain Tumor Detection
https://www.kaggle.com/navoneel/brain- Using Deep Neural Network and Machine
mri-images-for-brain-tumor-detection Learning Algorithm“International
[2]. Ed Burns “deep learning” [April Conference on Computer and Knowledge
2021]https://www.techtarget.com/searchente Engineering [October 2019]
rpriseai/definition/deep-
[15]. IlyasseAboussaleh “Brain Tumor
learningdeepneuralnetwork#:~:text=Deep%2 Segmentation Based on Deep Learning’s
0learning%20is%20a%20type,includes%20s
Feature Representation” Jounral of Imaging
tatistics%20and%20predictive%20modeling.
[December 2021]
[3]. Simplilearn “What is Perceptron: A [16]. Gajendra Raut “Deep Learning Approach
Beginners Guide for Perceptron” [February for Brain Tumor Detection and
2022]https://www.simplilearn.com/tutorials/ Segmentation” IEE [May 2021]
deep-learning-tutorial/perceptron [17]. G.Hemanth“DESIGN AND
[4]. Jason Brownlee “Transfer Learning in IMPLEMENTING BRAIN TUMOR
Keras with Computer Vision Models”[May DETECTION USING MACHINE
2019]https://machinelearningmastery.com/h LEARNING APPROACH” IEE [August
ow-to-use-transfer-learning-when-
2019].
developing-convolutional-neural-network-
models/
[5]. Sneha Grampurohit “BRAIN TUMOR
DETECTION USING DEEP LEARNING
MODELS”, IEEE [May 2020]

DOI: 10.35629/5252-04044145 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 45