2024 Second International Conference on Advanced Computing & Communication Technologies (ICACCTech)

DEEPTUMORNET: BRAIN TUMOR

CLASSIFICATION IN MEDICAL IMAGING
Mehak Sharma Krish Monga Ripandeep Kaur
Department of CSE Department of CSE Department of CSE
Chandigarh University Chandigarh University Chandigarh University
Mohali, India Mohali, India Mohali, India
[email protected] [email protected] [email protected]

Sahil Bhardwaj
Department of CSE
Chandigarh University
Mohali, India
[email protected]

Abstract— Brain tumors affect the survivability and the The adoption of CNNs for pathological tumor
quality of life of a patient and therefore the brain tumors issue classification in the medical field imaging technique has been
is a pressing international medical problem. Intervention must among the most trending health technologies in the recent
be effective. It should happen only when the early diagnosis of past. In order to implement the same, the need to invent
the disease is accurate and timely. The other side of medical computer-assisted diagnostic tools that will help radiologists
picture analysis is that it is labor-intensive and error-prone. The
to speedily and correctly identify brain tumor has advised us
perspectives of CNNs and medical imaging coupled with the
progress of statistical tools create a perfect platform for
to take this step. Along with this, they can perform the task of
improving the diagnostic accuracy of diseases. In the present image scanning in a span of short time which has the ability to
research, we put the use of CNNs for the brain tumor image automatically reduce the workload of the radiologists yet help
classification into the spotlight and we compare their upsides in carrying out the most accurate of diagnostics.
and downsides. We show that the CNN is capable of disrupting The human brain is affecting the two different types
the brain tumor examination and treatment planning by means tumors which are benign and malignant. Fortunately, the
of an extensive and in-depth investigation.
benign tumors are the least occurring tumor types with a slow
Keywords— Deep learning, Convolutional Neural Network,
but steady spread only as far as their site of origin without any
Brain Tumor, Medical Image Classification, Diagnosis Assistance chance of growing elsewhere. Whereas benign brain tumor is
made up of cells that originate from a specific part of the brain
I. INTRODUCTION and subsequently spread to other parts including the spinal
cord, the malignant brain tumor is composed of cancerous
Brain tumor, regardless of being malignant or benign, is
cells that begin in one area of the brain and quickly transfer
one of the troubling corresponding factors that affect normal
and grow into other regions. Nowadays, the brain tumors’
development of brain. The genesis of the diseases is most
evaluation and categorization depend on a visual analysis of
commonly a brain region, whilst sometimes they spread to
the biopsy samples. In contrast to this technique, a much easier
other part of the human body. In the United States, American
and less time-consuming approach is human error-prone
Brain Tumor Organization statistics report that there are
though. The automated deep learning system needs to be
87,000 newly diagnosed primary brain tumors every year. The
developed that for efficient detection and classification of
main keys to proficient neuro-oncology practice, proper
brain tumors. Thus, it would ensure the lack of currently
treatment, and a patient’s recovery are time-prompted and
existing constraints.
accurate diagnosis of brain tumors. Meanwhile, however, the
use of the conventional approach of evaluating medical
images by hand e.g. CT and MRI scans, takes a fair long time
and is very subjective, thereby, increasing the chance of error II. LITERATURE SURVEY
and reducing diagnostic accuracy. Besides increasing the precision in diagnosis and the
Being the last word, with time, and as medical imaging treatment effectiveness, one of the most interesting medical
technologies are improving, convolutional neural networks research areas is the application of convolutional neural
are generality turning out to be a winner in numerous Category networks, CNNs, for the classification of magnetic resonance
tasks images. CNNs feed on convolutional layers and use imaging patterns. Furthermore, CNNs have been manifested
deeper learning algorithms to characterize complicated recently as a suitable tool for tumor detection, automatic
patterns embedded in medical pictures. This is a piping interpretation and analysis of histological images in the frame
process which then utilizes the features within fully connected of medical diagnostic. This current study endeavours to
layers for categorization. Since, the large-dimensionality of inform about the methods, outcomes, and barriers encountered
medical data make the CNNs efficient for discriminating the when training deep learning-based CNN networks for brain
key features and the data patterns otherwise impossible to tumor detection and diagnosis.
detect with traditional techniques. In that regard, the CNNs are Stages of Initial Adoption and Deployment: The CNNs
of growing importance for medical image evaluation[1]. was novel to assist medical imagery interpretation as early in
2010s era. The data of medical imaging performed the role of
correcting the neural networks which were at the beginning

979-8-3315-1905-6/24/$31.00 ©2024 IEEE 500

DOI 10.1109/ICACCTech65084.2024.00087
prepared for general actions like image recognition. This 1) Support Vector Machines (SVMs): Among all
subsection hinges on the fact that CNNs were employed to available classification techniques, SVMs are often applied to
tasks such as fractional tumor segmentation from MRIs, lung differentiate different types of brain tumor because the method
nodule detection of X-rays, and cancer diagnostics from permits to work with high-dimensional data and build decision
mammograms. These were significant milestones. Despite its planes separating the tumor types of maximal margins.
seeming high potential as discovered by CNNs, unprocessed
transmissions used to be note because of some issues in 2) Random Forest (RF): RF is one of the most
interpretation, computing capacity, and collection of ensembles learning algorithms that use a multitude of decision
magnitude. trees for better classification. Therefore, it stands out topping
its competition for being efficient when dealing with large and
Improvements and Modifications: While publication of skewed data sets in the classification of brain tumours.
these models (VGGNet, GoogLeNet, and ResNet) in the
middle of 2010 offered an engineering breakthrough and 3) k-Nearest Neighbors (k-NN): k-NN is one of the
distinguished between the network architectures of the CNN, basic ushing the classification of tumor depends on the
the progress see was not unceasing and was instead very similarity an unknown sample has to the closest data point on
uneven. In the field of medical image processing investigators the training sets. It is suitable for cases where the data sets are
comparatively small and whenever, processing speed is an
started to use a method of training called transfer learning
where the networks that had been trained for healthcare issue.
purposes using little amount of data would then be easily 4) Genetic Algorithms (GA): As it has been illustrated,
modified so they could operate on the large- scale images GA imitates natural selection to pick the best choices that
datasets such as ImageNet. Using various imaging modalities would serve the classification stages properly. It is used
(i.e.MRI, CT, and PET) to generate more accurate image sometimes in conjunction with other classifiers to improve
fusion strategies has made comprehensive classification and accuracy when detecting tumor as far as eliminating false
diagnosis of many diseases possible in a more effective positives is concerned.
manner. Investigation was strived to even data enrichment,
class weighing, and the attention mechanisms to solve 5) Fuzzy Logic Systems: In the context of medical,
problems, data scarcity, class inequality among the students, especially imaging data, fuzzy logic is applied for brain
and model understanding[8]. tumors classification due to their uncertainty and imprecision.
This approach emulates a form of human-like reasoning to
Current Developments and Advances: During the last make decisions with partial truths, which is critically
decade there has been a wide increase of researches aimed at applicable in medical imaging.
making Convolutional Neural Networks to understand and
explain their role in the area of medical image analysis. This 6) Principal Component Analysis (PCA): PCA is used
is being done by creating network nodes, self-focus in dimensionality reduction; evaluating the most important
mechanisms, and comprehensible artificial intelligence (AI) features from large imaging datasets. This technique is often
techniques in both optical and predictive thinking processes. used optionally prior to classification with the aim of
This has thus improved physicians' faith and acceptability and increasing speed as well as the quality of results.
also the level of understanding of computerized 7) Deep Belief Networks (DBNs): The reviewed DBNs,
diagnostics[4]. It was proposed, the hostile learning methods which belong to the generative neural network family, are
to decrease CNN models' vulnerability to mischievous input, applied to discover hierarchical abstractions of medical
region shifts, and give constant precision across a wide images. Such networks assist in characterizing the brain
geometry of people's upload demographics and imaging tumour by learning the abstract features from the raw image
techniques. Recent studies and public databases that are data.
shared within the research community have enabled CNN
algorithms to become more reproducible and easier to validate 8) Autoencoders: Autoencoders are a type of neural
and so they can be adopted faster into the scientific field. network which have no input-output training pairs and which
try to transmit information through a bottleneck. They are
The distinguishable uniqueness of convolutional neural applied in feature extraction and reduction in the classification
networks (CNNs) emerges as a wonderful tool for brain tumor of brain tumors, which therefore improves the efficiency of
image classification; this will allow for a paradigm shift in the classifiers.
field that will enable earlier disease identification leading to
tailored treatments. While there is significant progress, further III. METHODOLOGY
research is still required for the solution of the problems This study paper aims to present an innovative network
existed today and invention of upgraded methods of design specifically made for tumor grading and sorting in
aggrandization of the reliability and efficacy of CNN-based magnetic resonance (MR) scans. The proposed network layout
diagnostic systems in the real life situations are all the wants. tries to improve current methods by giving more accurate
As the topic is right on point to the architecture aspects like tumor grades, working better, or both. A brief overview of the
instruction, comprehension and simulation for real-world full process is provided below, which is a picture showing the
applications, better usage of CNNs denotes the step to higher suggested technique or process involved. The different steps
engagement into the medical image classification space. To or parts of the sorting process, such as getting features,
leverage the limitless ability of CNNs and help in better grouping them, checking, and preparing data, are likely shown
healthcare delivery and optimal patient outcomes, the in this diagram.
departments and research projects are supposed to be
integrated[5].
Some of the advanced techniques used for brain tumor
classification in medical imaging are:

501
 Fig. 2. Images of types of Tumor

Fig. 1. Flowchart of the Process

A. DATASET
In this work, the dataset obtained for training and testing
purposes was obtained through public accessible databases Fig. 3. Rotated Images of Tumor
such as Kaggle[21]. This dataset encompasses pituitary,
glioma, meningioma, and no tumor as brain tumor categories.
The amount of photos in the dataset for every category is listed
in the following table. Every picture in the dataset has a
detailed description.
C. Proposed Method
TABLE I. IMGAGES USED IN DATASET
This study uses a Convolutional Neural Network (CNN)
technique to automatically identify brain tumors using health
Class Number of Images picture data. Using labelled input photos from raw data, the
study uses the intrinsic structures in these photos to distinguish
Glioma 826
among brain tissues that are benign as opposed to those that
Meningioma 822 have tumors. In particular, a collection of 2870 sample photos
is used for training the CNN model[14]. Of these, 2475 images
Pituitary 827
show several kinds of tumors, including meningiomas,
No Tumor 395 gliomas, and pituitary tumors, whereas the balance of 395
images show normal brain tissue. Through the use of this the
TOTAL 2870
information set, the CNN gains the ability to identify
distinctive features linked to the existence of tumors, which
helps it effectively categorize unidentified photos and support
an identification of brain tumors. By employing this strategy,
B. Convolutional Neural Network (CNN) this research hopes to further autonomous the processing of
CNNs are a kind of deep neural network that draw medical pictures approaches, which could lead to more swift
inspiration from the average person's brain's visual cortex. It and precise brain tumor identification and improved treatment
is made up of several layers that are in charge of acquiring and for patients[4].
retrieving attributes from images, such as convolutional
layers, pooling layers, and fully connected layers.
Convolutional layers process an initial image by applying
filters and looking for characteristics like boundaries and arcs.
The representation of features is easier to handle because of
the pooling layers, which shrink its size[3]. The image is
classified by the completely linked layers using the features
that were retrieved. Return propagation is used to train the
network, and the filter's weights are changed to minimisze the
discrepancy between the anticipated and observed picture
labels.
Determining medical pictures into categories such as
typical brain visuals or visuals exhibiting symptoms of a brain
tumor entails brain tumor image categorization. The intricacy
of brain pictures and the minute distinctions between a healthy Fig. 4. Proposed Method
and disordered brain make this effort difficult. For this
activity, CNN has proven to be an effective tool, attaining
outstanding precision and minimizing the requirement for
human interaction.

502
D. Performance Evaluation 2) Accuracy: The proportion of appropriately categorized
In order to gauge how well the advocated CNN design photographs relative to each photograph in the collection is
performs in categorizing brain tumors in both conventional referred to as correctness. Greater correctness indicates that
and enhanced pictures, a matrix of uncertainty must be the model appears more dependable and makes fewer
created. The accuracy, sensitivity, precision, and specificity of blunders. For a number of explanations, accuracy plays a
the model in classifying brain tumors may all be thoroughly critical role in CNN-based brain tumor categorization of
examined thanks to this matrix. The matrix offers perspectives images[11]. First and foremost, an excellent accuracy rating
on the model's virtues and drawbacks by methodically demonstrates the dependability and durability of the model,
analyzing the projected designations over its initial truth
indicating that it is appropriate for usage in real-world
labels[12]. This helps investigators to comprehend the model's
functionality across various brain tumor types and image situations. Minimizing the probabilities of misleading results
changes. In the end, this evaluation procedure will help to and erroneous negatives requires excellent precision. When a
improve and optimize mechanical brain tumor detection benign tumor (tumor that is not cancerous) is classified as
systems by offering a comprehensive analysis of CNN's malignant by the model, this is known as a mistaken
efficacy in tumor categorization. positive[15]. Conversely, erroneous negative results happen

when a malignant tumor is classified as benign by the model.

Fig. 5. Confusion matrix of our Model

1) Heatmap: Whenever it's necessary to comprehend the

CNN's choice- making procedure, heatmap brain tumor
picture categorization plays an integral role. The parts of the
photograph which significantly influence a categorization
determination are shown graphically in heatmaps. Stated
differently, the heatmaps indicate which regions of the
cerebral cortex MRI the CNN applies to as a basis for
categorisation[3]. This is crucial for illness identification
because it enables radiologists to comprehend the logic
underlying CNN's judgement and validate or invalidate it in Fig. 7. Accuracy
light of their very own experience. In addition, heatmaps
offer a rationalisation for the categorization outcomes, which
is critical for the medical profession to grasp and have faith
in CNN-based approaches to diagnosis. Additionally,
heatmap brain tumour picture categorization with CNNs
helps find concealed or inconspicuous tumors that could go
unnoticed by human professionals or even by conventional
approaches[5]. This has the potential to enhance patient
satisfaction and maybe prevent needless procedures by
aiding.

Fig. 8. Adjusted the convolution network's parameters to identify

divisions

IV. RESULTS AND ANALYSIS

Criteria for assessing performance: The hypothesis was
always that varied tools can show effectiveness in space.
Problems were caused by the confusion matrix that signalled
Fig. 6. Heatmap the difference in community density, the most accurate
perspectives, and observation error. Testing a model

503
demonstrated where it is strongest as well as its weakness to a
number of tumors types and a corresponding electronic
microscopic image[2]. The set of a line plot in the form of the
heat map distinctly showed whether some of the MRI scans
regions presumed to influence the classification process are
more critical than others. This aids in the detections of small
tumors precisely as well as the high-quality image contributed
to proper diagnosis uncertainly, which then might have lead to
a good patient outcome.
Fig. 9. Pituitary Tumor detected by the model
Utilizing the Proposed Approach: The proposed CNN
model has been thoroughly trained on a diverse set of dataset
having images which included healthy brain tissues and
varieties of brain tumor. These features have been associated
uniquely and exclusively with tumor images, giving the model
the ability to classify the category of the non-viewed photos
on the basis of such feature set[9]. During the completion of
the model effectiveness evaluation with comparison of both
regular and improved images, the classifier performance
might be spotted for misclassifying tumors as well as good
predictive tumor classifier performance[6]. Researchers came
to a better understanding of the positive and negative side of
the model by means of a purposeful analysis of specialty and Meningioma
Fig. 10. Tumor detected by the model
precision, accuracy, and sensitivity parameters. To determine
the distribution of brain cancer patients, thorough research
process was performed, with the ultimate goal being
investigation, refining and achieving of humanized automatic
brain tumor recognition software.
Medical imaging and computerized diagnostic techniques,
more specifically, brain tumor classification has great
importance in today’s advanced medical science. Of them, one
of the primary applications is the improvement of early
diagnosis capability; with imaging techniques, including MRI
and CT, and supported by machine learning, it can identify
different types of tumors. This is made easier due to
technological advancement hence making diagnoses to be
quicker and specific hence early interferences and treatment Fig. 11. No Tumor detected by the model
programs to be formulated. Also, in tending large databases
modern classifiers help to trace changes in tumors, to define
advancement in size, form, or other crucial characteristics,
therefore being irreplaceable in monitoring the treatment
efficacy, including surgery, radiation, or chemotherapy. They
also assist the clinical decision making process because they
use data analytics to provide prognosis of the patients and
ways of enhancing the clinical treatment plans to suit each
client. In clinical trials, tools for classifying brain tumors are
used mainly for objective assessment of the reaction of tumor
to different experimental treatments and for obtaining
measurable values that help in development of treatment
methods. Furthermore, through the use of suchsystems,
unnecessary reliance on manual interpretations reduces and Fig. 12. Giloma Tumor detected by the model
hence increase the correct diagnosis particularly intreating the
developing world where there is limited access t expert
radiologists. Finally, the possibilities to incorporate the given V. CONCLUSION
brain tumor classification into the medical imaging process
are proclaimed to posses a relatively high level of potential for All in all, the application of convolutional neural networks
improved results of patient treatment and further development (CNNs) to brain tumor categorization is certainly the most
of medical neuro-oncology studies. important step in medical diagnostics towards boosting the
sensitivity and efficacy of brain tumor recognition and
Following are the results that were detected by our model: classification process. This work is as a milestone to the area
of development of advanced methods in the sphere of
improved diagnosis of brain tumors
thorougly analysing previous articles and developning new
CNN-based method[7].

504
 The high level of tedious work and human prone error in Classification in MRI Images Using En- CNN”, International Journal of
traditional laborious images processing led to the emergence Intelligent Engineering and Systems, 14(4), 437–451, 2021
[6] Rasool, M., Ismail, N., Boulila, W., Ammar, A., Samma, H., Yafooz,
of the necessity of medical imaging aids which are accurate W. S., & Emara, A. H., “A Hybrid Deep Learning Model for Brain Tumour
and automated. The concept of Human Extra-Sensory Brain Classification”, Entropy, 24(6), 2022
(CNN) development inspired from human sensory system and [7] Khan, A. H., Abbas, S., Khan, M. A., Farooq, U., Khan,W. A., Siddiqui,
its effectiveness in processing medical images. The emerging S. Y., & Ahmad, A, “Intelligent Model for Brain Tumor Identification Using
AI systems are able to dig deep into the extra special structure Deep Learning”, Applied Computational Intelligence and Soft Computing,
2022
and features from the varied pool of health data. [8] Alsubai, S., Khan, H. U., Alqahtani, A., Sha, M., Abbas, S., &
The CNN proffers advanced classification procedure for Mohammad, U. G., “Ensemble deep learning for brain tumor detection.”,
Frontiers in Computational Neuroscience, 16, 2022
brain tumor scans using its extensive database which [9] Shajin F, P S, Rajesh P, Nagoji Rao V, “Efficient Framework for Brain
comprises of normal functioning brain issues and various Tumour Classification using Hierarchical Deep Learning Neural Network
brain tumors. The accuracy of the model in detecting Classifier”, Computer Methods in Biomechanics and Biomedical
cancerous growth like brain tumor is tested by employing Engineering: Imaginf and Visualization,11(3), 750–757, 2023
strict evaluation methods that include sensitivity, precision, [10] Sunanda Das, O.F.M. Riaz Rahman Aranya, Nishat Nayla Labiba,
“Brain Tumor Classification Using Convolutional Neural Network”,
accuracy, and specificity. This procedure will definitely Researchgate, May 2019
allows people to make an informed decision about the model [11] Ali Mohammad Alqudah, Hiam Alquraan , Isam Abu Qasmieh , Amin
by recognising its benefits and defects[13]. Furthermore, Alqudah , Wafaa Al-Sharu, “Brain Tumor Classification Using Deep
visualisation techniques like heatmaps are complex hence Learning Technique - A Comparison between Cropped, Uncropped, and
making it easy to comprehend the CNNs decision-making Segmented Lesion Images with Different Sizes”, International Journal of
process thereby alerting experts about possible diagnostic Advanced Trends in Computer Science and Engineering,
Volume 8, No.6, November – December 2019
errors whilst also having the information enables them to have [12] D C Febrianto1, I Soesanti1 , H A Nugroho, “Convolutional Neural
an insight into the basis of categorization by the model. It Network for Brain Tumor Detection”, 2nd International Conference on
ensures cancerous tumor of the brain is spotted at earliest and Engineering and Applied Sciences (2nd InCEAS), 2020
earliest point and cerebral tumors are diagnosed accurately, [13] Soheila Saeedi , Sorayya Rezayi, Hamidreza Keshavarz and Sharareh
which has a positive impact on the treatment of patients and R. Niakan Kalhori, “MRI-based brain tumor detection using convolutional
deep learning methods and chosen machine learning techniques”, BMC
instills confidence in the use of this method. Medical Informatics and Decision Making, 2023
Eventually, the outcomes of this research give the reason [14] R. Dhiman, G. Joshi, and C. Rama Krishna, “A deep learning approach
for Indian sign language gestures classification with different backgrounds”,
to think that CNNs are the true game changer in the sphere of Journal of Physics Conference Series 1950(1):012020,
medical image processing, especially because they are August 2021
remarkably successful in brain tumor diagnosing. On-going [15] Richa Dhiman, Pooja Luthra, Nongmeikapam Thoiba Singh, “Different
imperative and multidisciplinary co-operation between the Categories of Feature Extraction and Machine Learning Classification M [6]
technological and medical areas where new researches are Noreen N, Palaniappan S, Recognition Systems: A Review”, IEEE 8th
International Conference for Convergence in Technology (I2CT), April 2023
done is the fact to keep up with evolving issues and [16] Noreen N, Palaniappan S, Qayyum A, Ahmad I, Imran M, Shoaib M, “
enhancement of medical CNN-based diagnostic devices. With A Deep Learning Model Based on Concatenation Approach for the diagnosis
CNNs' abilities to catch on the disorder in the first place and of Brain Tumore”, IEEE Access, 2020
tailor individual treatments, we can realize these objectives to [17] Akkus, Z., Ali, I., Sedlar, J., & Agrawal, J. P., “Deep learning for brain
bring benefit to patients around the world. MRI segmentation: state of the art and future directions”, Journal of Digital
Imaging, 30(4), 449- 459, 2017
REFERENCES [18] Chang, P., Grinband, J., & Weinberg, B., “Deep- learning convolutional
neural networks accurately classify genetic mutations in gliomas”, American
[1] J. Seetha and S. Selvakumar Raja, “Brain Tumor Classification Using Journal of Neuroradiology, 39(7), 1201-1207, 2018
Convolutional Neural Networks”, Biomedical & Pharmacology Journal, [19] Noreen N, Palaniappan S, Qayyum A, Ahmad I, Imran M, Shoaib M.,
Vol. 11(3), September 2018. “A Deep Learning Model Based on Concatenation Approach for the
[2] Amou, M. A., Xia, K., Kamhi, S., & Mouhafid, M., “ A Novel MRI diagnosis of Brain Tumore”, IEEE Access, 2020.
Diagnosis Method for Brain Tumor Classification Based on CNN and [20] Soheila Saeedi, Sorayya Rezayi, Hamidreza Keshavarz & Sharareh R.
Bayesian Optimization” Healthcare (Switzerland), 10(3), 2022 Niakan Kalhori , “MRI-based brain tumor detection using convolutional
[3] Asif, S., Yi, W., Ain, Q. U., Hou, J., Yi, T., & Si, J.,“Improving deep learning methods and chosen machine learning techniques”, BMC
Effectiveness of Different Deep Transfer Learning-Based Models for Medical Informatics and Decision Making, January 2023
Detecting Brain Tumors From MR Images”, IEEE Access, 10, 34716– [21] M. Nickparvar, ‘Brain Tumor MRI Dataset’. Kaggle, 2021.
34730, 2022
[4] Swarup, C., Singh, K. U., Kumar, A., Pandey, S. K., varshney, N., &
Singh, T., “Brain tumor detection using CNN, AlexNet & GoogLeNet
ensembling learning approaches”,Electronic Research Archive, 31(5), 2900–
2924, 2023
[5] Tjahyaningtijas, H. P. A., Rumala, D. J., Angkoso, C. V., Fanani, N. Z.,
Santoso, J., Sensusiati, A. D., … Purnomo, M.H. , “Brain Tumor

505