International Journal of Engineering Trends and Applications (IJETA) – Volume 11 Issue 3 May - Jun 2024

RESEARCH ARTICLE OPEN ACCESS

A Review on Classification of Brain Tumor by Deep Learning

Using Convolutional Neural Network
Dr. Rahul Misra [1], Sudhanshu Vashistha[2]
[1]
Assistant Professor, Department of BCA, Agra College, Agra, U.P, India
[2]
Assistant Professor, Department of CSE, Jagannath University, Jaipur, Rajasthan, India

ABSTRACT
Brain tumors are classified using a biopsy, which is not normally done before conclusive brain surgery. The enhancement of
this technology by ML could aid radiologists in tumor detection without the use of invasive procedures. The convolutional
neural network is a ML-based algorithm that has achieved meaningful output and response in image segmentation and
classification. Pattern and image recognition problems commonly employ convolutional neural networks (CNNs). Automatic
brain tumor classification is a tough challenge because of the huge spatial and structural diversity of the brain tumor's
encompassing area. Convolutional Neural Networks are capable of performing this function with simplicity. In this paper, we
have performed a comparative study of different tumor classification methods using CNN.
Keywords — ML, tumor, Deep-learning, convolutionalneural network, image-classification, MRI imaging, Brain Tumor.

clinical discovery frameworks [9]. Many artificial intelligence

I. INTRODUCTION technologies have recently been used to identify and recognize
The In the last decades as indicated by the World Health brain tumors, including SVM, ANN, CNN [10-14]. CNN is
Organization, "Cancer or Malignant growth is the significant perhaps the most recent innovation and advanced in the
justification behind death internationally" [1]. As a general machine learning domain, and it is utilized in the analysis of
rule, mind growths are quite possibly the most widely infections utilizing clinical pictures, explicitly CT and MRI
recognized and forceful threatening cancer sickness, with a pictures. Since it does not require pre- processing or feature
high mortality rate if diagnosed at a higher grade [2]. As a extraction before the training process, CNN has recently
result, after detecting the tumor, brain tumor grading is a become useful in clinical pictures categorization and grading
crucial step in determining a successful treatment strategy [3- [15-16]. By and large, CNNs are utilized to manage crude
4]. Meningioma, glioma, and pituitary tumors are the most pictures and are intended to dispose of or counterbalance a
common types of brain tumors depending on the affected portion of the information pre-handling steps. The input layer,
region, and they are categorized as benign (noncancerous) and convolution layer, RELU layer, totally associated layer, order
malignant (cancerous). Malignant tumors in the brain may layer, and result layer are among the few piled layers that
quickly spread to other brain tissues, deteriorating the patient's make up CNN. The process of CNN comprises of two cycles:
condition [5]. Brain tumors are classified according to their the convolution, and the down inspecting [17].
position, type of tissues involved, whether they are cancerous
or non-cancerous, Origination place, such as primary or
secondary and so on. According to the World Health
Organization, there are 200 distinct types of brain tumors
based on cell origin and cell activity, ranging from mild to
aggressive [1]. Tumors are also divided into benign and
malignant. Due to the obvious shape, size, position, and
condition of tumor in the brain, its identification is extremely
difficult. It is difficult to detect brain tumors at the beginning
of their forming because it is impossible to reliably determine
the tumor's size and resolution [6]. If a tumor is identified and
cure at the beginning of the tumor development process, the
probability of a patient being treated is very high. As a result,
timely tumor diagnosis is important for effective treatment [7].
A medical examination, as well as computer tomography or
magnetic imaging, are usually used to make the diagnosis [8].
One of the primary and important techniques for detecting and
evaluating the patient's brain is MRI imaging, which produces Fig. 1. Types of Brain Tumor
accurate images of the brain. Due to their higher dissimilitude
in mushy tissue in persons, MRI pictures give improved II. LITERATURE REVIEW
results than other imaging strategies like CT in the field of

ISSN: 2393-9516 www.ijetajournal.org Page 38

International Journal of Engineering Trends and Applications (IJETA) – Volume 11 Issue 3 May - Jun 2024
The utilization of DL and ML methods to detect and grade classification method based on a convolution neural network.
brain tumors using a variety of imaging techniques, especially The outcomes of the classification are also classified as either
MRI, is now widespread. This part presents the most current tumor or non-tumor brain images. For classification purpose,
and associated research works which have written in various an image net database is used. This model has proactively
papers. been prepared. Subsequently, just the last layer is prepared.
In [16], Two approaches for classifying brain tumors have CNN likewise determines crude pixel values with width, level
been suggested by the authors. They used CNN in their first and profundity, highlight values. At last, to accomplish high
approach, which had four convolution layers, four pooling accuracy, the Gradient plunge based misfortune work is
layers, one completely connected layer, and some utilized. The accuracy of training's rate is 97.5 per cent.
intermediate layers for normalizing data. The accuracy of this Validation precision is also high, with low validation loss.
approach in classifying brain tumors is 81.09 per cent. They In [21], When all of the GLCM features are combined with
used CNN to extract features and KELM to identify them in Contrast, accuracy can be enhanced. As a result, contrast is
the second method. On this dataset, the accuracy of the KE- critical in assessing when brain tumors are identified. Two
CNN process is 93.68 per cent, indicating that it is more combined features with Contrast would be more accurate than
reliable than the previous one. On the Based-on Literature two combined features without Contrast. Although the
Review, we also come up with a comparison table in which accuracy results including the Contrast work in the order
the techniques, used to perform the classification methodology interaction on a blend of mind growth information are
are discussed. immaterial, Contrast will further develop exactness by 20% on
In [17], The researchers took five pre-trained models to all mixes of highlights including Contrast.
identify brain tumors: ResNet50, Xception, InceptionV3, In [22], For classifying three different types of brain tumors,
VGG16, and MobileNet. Glioma tumor, meningioma tumor, the authors suggested a new CNN architecture. The created
no tumor, and pituitary tumor were the four ranks in the model. network was viewed as rudimentary than prior pre-prepared
In this paper, the author aims to improve the effectiveness of networks on T1-weighted contrast-improved attractive
MRI machines in classifying and detecting brain tumors. They reverberation pictures. The organization's accomplishment
validated five pre-trained design with pictures after training was estimated utilizing four methodologies: two information
and validating them. They were able to achieve accuracies of bases and a mix of two 10-overlay cross-approval strategies.
between (97.25 per cent and 98.75 per cent). With a 98.75 per Subject-wise cross-approval, one of the 10-overlap draws near,
cent accuracy score, the Xception model was the most was utilized to decide the organization's speculation capacity,
accurate. and the improvement was tried utilizing an expanded picture
In [18], based on DNN, a fully automated brain tumor information base. The record wise cross endorsement for the
classification system has been proposed. Images of low-grade expanded enlightening file yielded the best result for the 10-
and high-grade glioblastoma were also used to construct the overlay cross endorsement process, with an exactness of
proposed networks. This article represents a new CNN 96.56%.
structure. In this cascading architecture, the yield of a vital In [23], Using a Deep Neural Network classifier, the
CNN is utilized as a supplemental origin of knowledge for the researchers divided 66 brain MRIs into four main categories:
upcoming CNN. Image processing algorithms and techniques metastatic bronchogenic carcinoma, normal, glioblastoma,
are extremely useful in this area of research and offer a second sarcoma tumors.
opinion to help radiologists enhance their diagnosis and In [24], The authors of this paper suggest a new method for
diagnostic accuracy. categorizing brain medical images into normal and abnormal
In [19], For brain tumor segmentation, the author presented groups based on ConvNet. Brain tumor images are often
a CNN based classifier technique. The algorithm has two divided into two categories: low and high. As a conceptual
stages that are connected. In the primary phase, they trained tumor classification technique, they used an altered version of
CNN to learn the mapping from image space to tumor label the profound learning design on attractive reverberation
space. They sent the predicted label response from CNN, pictures. Since the preparation set names are at the picture
along with the testing image, to an SVM classifier for accurate level on the other hand than the pixel level, the order is
segmentation during the testing process. The deep CNN-SVM performed on the whole picture. The outcomes showed
classifier is again repeated. Experiments and comparisons incredible progress in describing cerebrum pictures, with a
showed that the proposed system performs admirably either precision of 91.16 percent.
CNN-based or SVM-based segmentation independently. As In [25], the research is intended to help doctors detect brain
compared to other techniques, the proposed method performs tumor utilizing MRI pictures. In this analysis, images from
admirably. It also outperformed CNN or SVM alone in terms The Cancer Imaging Repository (TCIA) were used. On
of tumor classification by a significant margin. images with a tumor, the suggested method has a recognition
In [20], Fuzzy C Means (FCM) based segmentation is rate of 94.28 per cent, and on images without a tumor, it has a
carried out Standard brain tumor classification. The level of recognition rate of 100 per cent. Their system has a 96 per
difficulty is poor. However, the computation time is long, and cent overall success rate, which is a better result than the
the precision is poor. To increase accuracy and reduce comparison.
computation time, the proposed scheme utilizes a

ISSN: 2393-9516 www.ijetajournal.org Page 39

International Journal of Engineering Trends and Applications (IJETA) – Volume 11 Issue 3 May - Jun 2024
Table 1 : Comparative Analysis of various methods for detection and classification of brain tumor

Ref. Data Set Use Techniques Accuracy/Result Limitation

Two-pathway
A novel two-path way Create a separate pre
CNN architecture
[18] 2013BRATS architecture is used to diction for each
and Cascaded
achieve high performance. segmentation label.
architectures

For brain tumor

As the window sizes for
segmentation, a
Independently greater than SVM is increased the
[19] MRIimagesof30patients CNN-based
CNN-based segmentation. precision of the samples
classifier model
will decrease
was adopted

[20] CNN Based

Training is performed
BRATS2015 Classification 97.5%
for the only final layer.
Method

[21] The contrast feature is

Mg-pt,Mg-Gl, Gl-Pt,
CNN with GLCM 82% significant for gene
and Mg-Gl-Pt
rating higher accuracy

98.75%, 98.50%, 98.00%,

Pre-trained models are
[17] From Kaggle web site Five pre-trained CNN 97.50%,
architectures
and
required
97.25%respectively

NHG Dedicated hardware for

[22] CNN architecture 96.56%.
Hospital, China mealtime performance.

Harvard Medical When compared to

High accuracy possible
[23] DWT with DNN conventional classifiers, high
School website with DWT
accuracy.

Only axial FLAIR-

weighted MR images
[24] (TCIA) CNN 91.16%.
have been used in this
study.

In accurately classified
tumor images and the
[25] TCIA Threshold method 96%
proposed method is not
automatic procedure.

T1-weightedcontrast-
[16] KE-CNN 93.68% Data set limitations
enhancedMR
is inferred that Deep learning offers the influential structure
for a brain tumor and which will provide considerable results.
III. CONCLUSIONS
This paper presents a comparative analysis of MRI-based REFERENCES
brain tumor classification techniques. The intention of this [1] World Health Organization—Cancer. Available online:
paper is to define the various methods for classifying brain https://www.who.int/news-room/fact-sheets/detail/
tumors using Deep Learning This comparative study reveals cancer. (accessed on 5 November 2019).
that each technique has advantages and disadvantages and that [2] G. K. Soni, A. Rawat, D. Yadav and A. Kumar, "2.4
they all yield important results in terms of performance GHz Antenna Design for Tumor Detection on Flexible
measurements. In this paper discussed the various techniques Substrate for On-body Biomedical Application," 2021
of brain tumor detection and classification. From this review it

ISSN: 2393-9516 www.ijetajournal.org Page 40

International Journal of Engineering Trends and Applications (IJETA) – Volume 11 Issue 3 May - Jun 2024
IEEE Indian Conference on Antennas and Propagation [15] Mathur, Neha, et al., “Detection of Brain Tumor in MRI
(InCAP), pp. 136-139, 2021. Image through Fuzzy-Based Approach”, High-
[3] G. Shankar, G. K. Soni, B. Kumar Singh and B. B. Jain, Resolution Neuroimaging-Basic Physical Principles and
"Tunable Low Voltage Low Power Operational Clinical Applications, Intech Open, 2018.
Transconductance Amplifier For Biomedical [16] Pashaei, A., Sajedi, H., & Jazayeri, N. “Brain tumor
Application," 2021 Fourth International Conference on classification via convolutional neural network and
Electrical, Computer and Communication Technologies extreme learning machines”, IEEE 8th International
(ICECCT), Erode, India, 2021, pp. 1-6, doi: conference on computer and knowledge engineering
10.1109/ICECCT52121.2021.9616627. (ICCKE), pp. 314-319, 2018.
[4] G. K. Soni, D. Yadav and A. Kumar, "Enhancing [17] Saleh, A., Sukaik, R., & Abu-Naser, S. S., “ Brain
Healthcare: Flexible and Wearable Antenna Design for Tumor Classification Using Deep Learning”, IEEE
Tumor Detection", IEEE 2024 International Conference International Conference on Assistive and Rehabilitation
on Distributed Computing and Optimization Techniques Technologies (iCareTech), pp. 131-136, 2020.
(ICDCOT), pp. 1-5, 2024. [18] M. Havaei, et al, Brain tumor segmentation with deep
[5] M. Karuna and A. Joshi, “"Automatic detection and neural networks, Medical image analysis, 35, 18-31,
severity analysis of brain tumors using gui in matlab”, 2017
International Journal of Research in Engineering and [19] B. Cui, M. Xie and C. Wang, "A Deep Convolutional
Technology, 10, pp. 586-594, 2013. Neural Network Learning Transfer to SVM-Based
[6] KS. Aboody, A. Brown, et al, “Neural stem cells display Segmentation Method for Brain Tumor," 2019 IEEE
extensive tropism for pathology in adult brain Evidence 11th International Conference on Advanced Infocomm
from intracranial gliomas”, Proceedings of the National Technology (ICAIT), pp. 1-5, 2019.
Academy of Sciences, 97 (23), pp. 12846- 12851, 2000. [20] Seetha, J., and S. S. Raja, “Brain Tumor Classification
[7] A. joshi, D. H. Shah, et al, “Survey of brain tumor Using Convolutional Neural Networks”, Biomedical &
detection techniques through MRI images”, International Pharmacology Journal, Vol 11, pp 1457-1461, 2018.
Journal of Research in Engineering and Technology, 10, [21] Widhiarso, W., Yohannes, Y., & Prakarsah, C., “Brain
pp. 586-594, 2013. tumor classification using gray level co-occurrence
[8] G. K. Soni, A. Rawat, S. Jain and S. K. Sharma, "A matrix and convolutional neural network”, IJEIS
Pixel-Based Digital Medical Images Protection Using (Indonesian Journal of Electronics and Instrumentation
Genetic Algorithm with LSB Watermark Technique", Systems), 8(2), pp. 179-190, 2018.
Springer Smart Systems and IoT: Innovations in [22] Badza, M. M., & Barjaktarovic, M. C. , “Classification
Computing, pp. 483-492, 2020. of brain tumors from MRI images using a convolutional
[9] JP. Poonam, “Review of image processing techniques neural network”, Applied Sciences, 10(6), 2020.
for automatic detection of tumor in the human brain”, [23] Mohsen, H., El-Dahshan, E. S. A., El-Horbaty, E. S. M.,
International Journal of Computer Science and Mobile & Salem, A. B. M., “Classification using deep learning
Computing, 2(11), pp. 117-122, 2013. neural networks for brain tumors”, Future Computing
[10] Dr. Himanshu Aora, Kiran Ahuja, Himanshu Sharma, and Informatics Journal, 3(1), pp. 68-71, 2018.
Kartik Goyal, Gyanendra Kumar "Artificial Intelligence [24] Khawaldeh, S., Pervaiz, U., Rafiq, A., & Alkhawaldeh,
and Machine Learning in Game Development", Turkish R. S., “Noninvasive grading of glioma tumor using
Online Journal of Qualitative Inquiry (TOJQI), vol-12, magnetic resonance imaging with convolutional neural
issue-8, pp. 1153-1158, 2021. networks”, Applied Sciences, 8(1), 27, 2018.
[11] Kiran Ahuja, Harsh Sekhawat and Shilpi Mishra, [25] Ilhan, U., & Ilhan, A., “Brain tumor segmentation based
Pradeep Jha“Machine Learning in Artificial Intelligence: on a new threshold approach”, Procedia computer
Towards a Common Understanding", Turkish Online science, 120, pp. 580-587, 2017.
Journal of Qualitative Inquiry (TOJQI), Volume 12, [26] B. Verma, S. Saraswat, V. Saraswat and R. Misra,
Issue 8, pp-1143-1152, July 2021. "Interstitial Lung Disease Patterns Classification using
[12] P. Jha, T. Biswas, U. Sagar and K. Ahuja, "Prediction Hybrid Features Set and Multi Level Segmentation
with ML paradigm in Healthcare System," 2021 Second Implemented by Machine Learning Algorithm", IEEE
International Conference on Electronics and Sustainable 2023 8th International Conference on Communication
Communication Systems (ICESC), pp. 1334-1342, 2021. and Electronics Systems (ICCES), pp. 1811-1815, 2023.
[13] Fabelo, Himar, et al. , “An intraoperative visualization [27] Rahul Misra, Dr. Ramkrishan Sahay, "A Review on
system using hyperspectral imaging to aid in brain tumor Student Performance Predication Using Data Mining
delineation”, Sensors, Vol 18, 2018. Approach", International Journal of Recent Research and
[14] Sasikala, M., and N. Kumaravel. A wavelet-based Review, Volume-X, Issue 4, pp. 45-47, 2017.
optimal texture feature set for classification of brain [28] Priya Gaur, Sudhanshu Vashistha, Pradeep Jha,
tumours. Journal of medical engineering & technology, "Comprehensive Review of Sentiment Analysis Using
Vol 32, pp198-205, 2008. Machine Learning", International Journal of Engineering

ISSN: 2393-9516 www.ijetajournal.org Page 41

International Journal of Engineering Trends and Applications (IJETA) – Volume 11 Issue 3 May - Jun 2024
Science Technology And Research (IJESTR), Vol. 7,
Issue. 3, pp. 37-46, 2022.
[29] P. Gaur, S. Vashistha and P. Jha, "Twitter Sentiment
Analysis Using Naive Bayes-Based Machine Learning
Technique", Sentiment Analysis and Deep Learning.
Advances in Intelligent Systems and Computing, pp.
367-376, 2023.
[30] Rahul Misra, Dr. Ramkrishan Sahay, "Evaluation of
Five-Class Student Model based on Hybrid Feature
Subsets", International Journal of Recent Research and
Review, Volume-XI, Issue 1, pp. 80-86, 2018.
[31] Babita Jain, Gaurav Soni, Shruti Thapar, M Rao, “A
Review on Routing Protocol of MANET with its
Characteristics, Applications and Issues”, International
Journal of Early Childhood Special Education, Vol. 14,
Issue. 5, pp. 2950-2956, 2022.
[32] Rahul Misra, Dr. Ramkrishan Sahay, "Evaluation of
Student Performance Prediction Models with TwoClass
Using Data Mining Approach ", International Journal of
Recent Research and Review, Volume-XI, Issue 1, pp.
71-79, 2018.
[33] A. Dhoka, S. Pachauri, C. Nigam and S. Chouhan,
"Machine Learning and Speech Analysis Framework for
Protecting Children against Harmful Online Content,"
2023 Second International Conference on Electronics
and Renewable Systems (ICEARS), pp. 1420-1424,
2023.

ISSN: 2393-9516 www.ijetajournal.org Page 42