2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)

Comparative Analysis of Deep Learning and

Machine Learning for Detection and Classification
of Brain Tumors at Multiple Stages
2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE) | 979-8-3503-2820-2/24/$31.00 ©2024 IEEE | DOI: 10.1109/IC-ETITE58242.2024.10493717

Mr. B.Ramprakash1 Mr. S.Santhana Hari2 Mr. D.Nagendra Kumar3

Assistant Professor, Dept. of CSBS Assistant Professor, Dept. of CSE Assistant Professor, Dept. of CSE
Thigarajar College of Engineering Thigarajar College of Engineering Thigarajar College of Engineering
(autonomous), Madurai – 15. (autonomous), Madurai – 15. (autonomous), Madurai – 15.
[email protected] [email protected] [email protected]

Ms. C.Santhiya4
Assistant Professor, Dept. of CSE
Thigarajar College of Engineering
(autonomous), Madurai – 15.
[email protected]

Abstract— The brain acts as the body's command central for such a connection to be made. It goes without saying that
location. As time goes on, more and more brain illnesses are the diagnosis of neurological illnesses is among the most
pinpointed. Because of this diversity, improving current pressing uses of BCI now under investigation. Because of the
methods of diagnosing or detecting brain illnesses is a constant brain's complex structure, which changes with age and
focus of study. The effectiveness of treatment for disorders of pathological history, detecting neuro-degenerative illnesses
the brain can be greatly improved by catching them early. has always been difficult. Detecting health problems early on
Recent years have seen the widespread adoption of AI across the is essential. Computer-assisted techniques are becoming
scientific community, and without a doubt, this has had a increasingly important in the diagnosis of many illnesses of
profound impact on neurology. Brain disease diagnosis and
the brain, replacing the more traditional manual methods. The
prognosis have both benefited from AI's incorporation into the
medical industry. Cancer of the brain is a major killer in the
brain is a complex organ made up of billions of individual
world today. Because of these characteristics, there are many cells. Brain tumours result from the uncontrolled
ways to check for brain cancer. Prompt diagnosis is essential for multiplication of these cells in or near the brain. Unregulated
effective treatment of brain tumors. One such technique is cell division in this population has been shown to disrupt
magnetic resonance imaging. In contrast, state-of-the-art normal brain and body functions. Depending on their size,
approaches have been employed to address several shape, and texture, brain tumours are either benign (low grade)
classification-related challenges in medical imaging in recent or malignant (high grade).
years. These include deep learning, neural networks, and
Automatic tumour classification and early cancer
machine learning. To differentiate between glioma,
meningioma, pituitary, and no tumour in patients with brain
diagnosis are essential for helping clinicians design effective
cancer, the SVM classifier in machine learning (ML) was cancer treatments. Techniques like computed tomography and
utilized in this study. The information in this study comes from magnetic resonance imaging can aid in the diagnosis of brain
people's contrast-enhanced MRI images. This study provides a cancer. Magnetic resonance imaging is a common treatment
comparison between the suggested model and competing models option because it does not involve any discomfort for the
to prove the superiority of our method. Both the raw data and patient and does not require radiation. In addition, because
the data after it had been cleaned and supplemented were MRI can create such high-quality images of brain tissue, it is
evaluated. In this research, we explore the possibility of using widely considered the most effective and widely used
deep learning and other advanced machine learning approaches approach for identifying and classifying brain tumours. In
to spot brain abnormalities. To be noteworthy concerns with order to find issues, specialists must manually evaluate
machine learning/deep learning-based methodologies for multiple MR images at once, which is a time-consuming and
identifying brain illnesses are discussed, and the most laborious process. Artificial intelligence technology has
noteworthy findings from the publications we analyzed are become increasingly important in recent years as a means of
presented. The ultimate purpose of the research is to determine averting this disaster. As artificial intelligence (AI) develops,
the best method for diagnosing various forms of brain diseases. it is increasingly being employed in tandem with computer-
aided diagnostic tools. Brain tumors and cancer are only two
Keywords— Brain tumour; Machine learning; Deep learning;
of the many disorders that can be quickly and accurately
MRI; SVM; CNN.
diagnosed with the help of CAD technology. First, typical
I. INTRODUCTION CAD systems analyse images to locate and segment lesions,
then they analyse tumors using numerical parameters to
Due to its vast potential in areas as diverse as brain extract characteristics, and finally, they depend on an
fingerprinting, disease diagnosis and prevention, adaptive e- appropriate machine learning technique to predict
learning, monitoring for states of fatigue, stress, and abnormalities.
depression, etc., brain-computer interface (BCI) research has
exploded in popularity over the past few decades. Connecting Multiple new sectors have begun using ML-based smart
the brain to an external device is made easier using BCI, which system applications. In order for these systems to be useful, it
does so by recording the single most important trait necessary is necessary to discover or extract relevant properties. When it

979-8-3503-2820-2/$31.00 ©2024 IEEE 1

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.
 2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)
comes to updating computer programs, deep learning is a benchmarks while doing performance analysis on our
cutting-edge specialization. Its design is made up of several networks. For the purpose of classifying and diagnosing brain
nonlinear layers, each of which improves on the accuracy with tumors, a range of knowledgeable individuals, such as
which it collects characteristics by feeding in the output of the Komarasamy & Archana (2023), have developed a number of
one below it. Diseases are diagnosed from MRI images using useful methodologies. The following paragraphs will
cutting-edge machine learning technology, namely elaborate on these procedures. In terms of the amount of time
convolutional neural network algorithms. They have also been needed for detection, the degree of accuracy required, and the
used for image processing among other medical applications. size of the tumor, there are currently a number of challenges
Since CNN does not require pre-processing or feature that need to be overcome. The earlier a brain tumor is found,
extraction prior to the training phase, it is frequently used to the more therapy options a patient has, which in turn increases
classify and grade medical images. Classifying images as the patient's probability of surviving the disease. A number of
normal or abnormal is the initial step in ML- and DL-based preprocessing and data augmentation procedures have been
techniques to diagnosing brain tumours. Next, abnormal brain created in order to improve the accuracy of classification. One
MR images are recognized according to distinct types of brain such procedure is the development of suggestions for skewed
malignancies. data (Asad et al., 2023), to cite just one example. The
succession of U-Nets was supposed to assist in the diagnosis
As samples, some pieces of contemporary literature are of cancer. Before brain tumors were segmented, this method
included below. When classifying brain MRI data sets for the was utilized to locate them in the patient's brain. The model
three different types of tumours (meningioma tumours, that has been suggested.
pituitary gland tumours, and glioma tumours), multiple CNN
deep learning architectures were utilized. These architectures The researchers Ramtekkar et al. (2023) present a method
included GoogleNet, AlexNet, and VGGNet. They were for the detection of brain tumours that is fresh, improved, and
successful in increasing accuracy to 98.69% by utilizing the extremely precise. The system makes use of a number of
VGG16 framework. To introduce a capsule network different methodologies, some of which are pre-processing,
(CapsNet) for identifying types of brain tumours. They also segmentation, feature extraction, optimization, and detection.
combined CapsNet feature maps from several convolution The pre-processing system makes use of many filters,
layers to boost accuracy performance. According to the final including the Gaussian, mean, and median filters, among
tally, we successfully classified 86.50% of data. AlexNet, a others. For picture segmentation, the threshold approach and
variant of a convolutional neural network, was utilized to the histogram method are utilized. The detection of brain
create a procedure for identifying glioma brain cancers. They tumors with the use of medical image processing is still
managed an impressive 91.16% precision using whole-brain difficult. Saladi et al., (2023) analysis to devise a technique
MR imaging. Deep CNN, which is built on convolutional that is both more accurate and specific in determining whether
neural networks, was proposed as a means to detect and or not a new born has a brain tumor. Because the brain of a
categorize malignant brain tumors. When it comes to brain baby is different from the brain of an adult in a number of
segmentation, Fuzzy C-Means (FCM) is the approach of significant ways, pre-processing techniques are helpful
choice. The aggregated data shows that the application has a because, in the absence of these approaches, it is more
97.5% rate of success. In addition, the principal component probable that the results will be erroneous
analysis and fuzzy k-mean technique segment the brain
tumour diseases so as to expedite the evaluation. As a possible To achieve accurate categorization, it is necessary to first
answer, a methodology was proposed which makes use of extract relevant attributes (Doshi et al., 2023) from the data.
both DWT and DL methods. By the end of the day, they had The probabilistic FCM technique is used in this study to
achieved a 96.97% success rate. In order to develop a system improve segmentation. This study lays out a plan for
for classifying brain tumours, researchers made use of the simplifying the high-dimensional MRI brain image (Panigrahi
convolutional neural network architecture as well as the gray- & Subasi, 2023) so that specific regions of interest can be
level conformation matrix. After viewing each image at 0, 45, revealed. Treating patients with brain tumours requires their
90, and 135 degrees, four characteristics were discovered: prompt diagnosis. Manual identification of brain tumours is a
risky and invasive process. Because of this, improved methods
energy, correlation, contrast, and homogeneity. The findings
of the study indicated that the hypotheses made were correct of medical imaging have been developed, such as magnetic
82.27 percent of the time. Through the examination of many resonance imaging , which is essential for the early diagnosis
substances, the creators of this research want to devise a of brain tumours. According to ZainEldin et al.'s research
computer-aided technique for the diagnosis of cancer. Within from 2022, the proposed method achieves an accuracy of
the parameters of this plan, images of brain tumors are 96.8% for our model. It can take some time to discover a brain
collected, pre-processed to reduce the amount of background tumour, making the knowledge and experience of the
noise, run through the severity of the tumor. radiologist all the more important. According to Kandimalla
et al.'s research from 2023, as the number of patients has
II. LITERATURE SURVEY increased, so has the amount of data that needs to be handled.
This has resulted in older methods being not only expensive
A. Evaluation of Related Work but also ineffective. The primary objective is to devise a
Siar & Teshnehlab (2019) investigated a deep learning workable strategy for utilizing MRIs to diagnose brain
techniques that could identify tumors based on images taken tumours so that decisions on the circumstances of individual
from MRI scans of the brain. The images were used in the patients can be made in a timely manner that is both efficient
training process. The first news network to regularly and accurate. On the Kaggle dataset, which was acquired from
incorporate visuals was CNN. The entirely connected softmax BRATS 2015 for the purpose of diagnosing brain tumours
layer was able to achieve the amazing accuracy of 98% in using MRI scans, our suggested method is tested. This dataset
terms of categorization. In addition to sensitivity, specificity, contains 3,700 MRI brain photographs, with 3,300 of those
and precision requirements, we also make use of accuracy pictures exhibiting tumours

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.
 2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)
B. Comparative Analysis being male. 1892 images were acquired from a total of 153
patients, 1666 for training data and 226 for testing. The
TABLE I. COMPARISON collected images were initially 512 pixels by 512 pixels.
B. Background Knowledge on ML And DL
To solve a problem, ML involves instructing a machine to
draw on its stored knowledge. There is a lot of excitement
about the prospect of using machine learning to solve
problems faster than people, especially given the low cost of
computer power and memory today. This enables the
processing and analysis of enormous amounts of data,
allowing for the discovery of insights and correlations that
would otherwise be invisible to the naked eye. The algorithms
that underpin its smart behaviour are designed to help the
machine generalize from its own experience and hence make
more accurate assessments. To be more precise, deep learning
refers to a set of methods for processing brain data that are
based on autonomous feature engineering processes.
Accuracy and efficiency are achieved by automatic feature
acquisition. The feature extraction technique used by the
Table 1 provides a consistent comparison of all of the prior classifier is the main dividing line between ML and DL.
research papers in this area, summarizing their key features in Classification performance in DL is significantly better than
a nutshell. You may get a quick overview of the methods used in ML due to the features gathered from its multiple non-linear
thus far by looking at Table 1. The results and their hidden layers. Figure 1 illustrates the key distinctions between
accompanying caveats have been highlighted for a more ML and DL.
thorough review.
C. Proposed Research Methodology
The outcome revealed that precision continued to be the
foremost concern of the specialists. As opposed to this,
however, the accuracy of MRI scans would not be of much
assistance due to the imbalanced class issue that is well
known. In the presence of unbalanced data, numerous ML and
DL techniques are accessible for enhancing model
performance. Accuracy and recall perform admirably when a
class discrepancy exists. In addition to veracity, the majority
of the research employed these. Accuracy is shown by the
former term, whereas recall shows whether or not the minority
class or guiding values were discussed. Both indicators hold
equivalent significance with regard to accuracy, from this
standpoint.
C. Proposed Model's Qualities and Uniqueness
The study's novel aspect is the novel solution proposed by
the researchers to aid in brain tumor detection in the collected
MRI picture, which offers high levels of accuracy. The Fig. 1. Comparative analysis framework for ML & DL - brain tumour
proposed model uses a method designed to provide training detection
for models with greater optimization than methods previously
used in the literature. Our model's utility for both researchers In the case of brain tumors, a dataset would most likely
and practitioners is underscored by the fact that it can be consist of medical imaging like MRI in addition to patient
trained with less computational resources. The research model information like age, sex, and medical history. Labels or
integrated machine learning with deep learning models to annotations that point out the location and kind of tumor that
reach a remarkable accuracy rate that would outperform may be seen in the photos might also be included in the data.
existing options. The collecting of data could be put to use either in the research
of the characteristics shared by different types of brain tumors
III. METHODOLOGY or in the training of machine learning algorithms for the
identification and categorization of tumors. Image registration
A. Dataset is able to accomplish the task of bringing together into a single
Brain MRI images of 153 patients with migraines who coordinate system pictures of the same patient that were
were referred to imaging centres include normal and brain obtained at different times or with various imaging modalities.
tumour patients. Following examination and diagnosis by a The purpose of the research was to develop an automated
physician, the collected images included images of the brains system for spotting brain malignancies in MRI images. Using
of eighty healthy patients. Include 1321 images, consisting of brain MRI data, our proposed method detects brain cancers
56 for testing data and 515 for train data. 73 malignancies using CNN and SVM. According to the reference, while
Include 571 images, with 170 images serving as test data and identifying and categorizing brain tumor MRI with CNN,
1151 as train data. Brain tumour patients range in age from 8 other optimization strategies, such as proceeding with an
to 66 years old, with 86 patients being female and 68 patients ensemble method, can be considered as competitors. The

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.
 2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)
proposed framework was broken down into multiple parts in images that have been mistakenly categorized. Both of these
light of the deep learning models references. The MRI scan of abbreviations are referred to collectively as "true positives"
the brain was the main input image. Noise was reduced by the and "true negatives," respectively. As shown in Figure 2,
use of data operations like thresholding and refractive error classes that were successfully selected and classes that were
correction. The MRI image database of the brain was studied wrongly picked are denoted by the terms false acceptance and
and enhanced. After that, the pictures were scaled down so false rejection, respectively
they could be included in the model. Figure 1 depicts the
detection outcomes of two distinct approaches. At first, we
will use a single machine learning technique for positive and
negative tumor detection in the brain. Second, the accuracy of
the two-class classification was enhanced by using a
convolution layer. There are four methods presented here that
work effectively for detecting brain tumors.
D. Feature Extraction Comparison of Ml & Dl Approaches
Machine learning and image processing use the original
dataset to build learning-supporting features. In cases where
an algorithm's input data is too vast, it is possible to transform
the data into a more manageable set of features. The term
"feature extraction" is used to describe the process of selecting
a subset of the most important features. When an agent is
represented by a subset of data rather than the full original
data, this is because the defined characteristics convey
information about the input data. Feature extraction is widely
used in the field of image processing to isolate the specific
parts or characteristics of a digital image that are of interest.
E. Detection Procedure for Brain Tumors
Scenario for brain tumours

Input: MRI pictures of the brain

Outputs: Determining images and identifying tumours in the
brain
Estimating the presence of a brain tumour using CNN Layer
(DL) and SVM
Pre-Process = Estimating noise and making the images
appear clearer
Split the data into sets of traing & test
if finding Normal (No Tumour)
stop Fig. 2. Confusion Matrix
else Brain Tumour (Pos or Neg – ML) & ( Pituitary or
Glioma or Meningioma or None – DL) B. Statistical Analysis
end if The numbers 0.2, 0.4, 0.6, 0.8, and 1.0 represent the values
on the radial axis of the radial chart. The radial chart is a
The study was centred around the utilization of graphical representation of the metrics (Sensitivity,
Convolutional Neural Networks (CNN) and Machine Specificity, Accuracy, and Precision) for two classes ('No
Learning (ML) classifiers in order to distinguish between Tumor' and 'Positive Tumor'). 0.2: This value represents the
different types of brain tumours for the purpose of prognosis. lower end of the scale, indicating a lower performance level
The classifiers underwent training using MRI scans that were for the metric. 0.4: This value is higher than 0.2 and indicates
tagged, and the tumours were afterwards annotated. The a moderate performance level for the metric. 0.6: This value
Convolutional Neural Network (CNN) was employed for the is higher than 0.4 and indicates a reasonably good
purpose of analysing Magnetic Resonance Imaging (MRI) performance level for the metric. 0.8: This value is higher than
pictures and making predictions regarding the presence and 0.6 and indicates a high performance level for the metric. 1.0:
location of tumours. An alternative methodology entailed the This value represents the upper end of the scale, indicating a
utilization of a two-dimensional convolutional neural network perfect performance level for the metric. The radial chart
to assess computed tomography scans of the brain with the allows you to compare the performance of the model for both
objective of identifying and categorizing instances of brain classes visually. Each class's data points on the radial chart
malignancies. show how well the model performs in terms of Sensitivity,
Specificity, Accuracy, and Precision. Higher values on the
IV. EXPERIMENT ANALYSIS AND DISCUSSION radial chart indicate better performance for the corresponding
A. Performance Analysis metric. In a radial chart, the values of each metric are plotted
on separate axes, and each class is represented as a data point
In this context, the abbreviation TP refers to true positives on the corresponding axis. The metrics' scale is uniformly
that are reliably discovered among glaucoma images, and the spaced from 0.0 to 1.0 on the radial axis. Figure 3 shows layer
abbreviation TN refers to true negatives that correctly identify validation findings. The segmentation graph demonstrates

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.
 2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)
how the layer was separated into accuracy and loss versus
epoch for better data analysis.

TABLE II. STATISTICAL MEASURES OF VARIOUS METRICS IN BRAIN

TUMOUR

Fig. 5. Tumor Detection in DL Classifier

Deep learning focuses on teaching artificial neural

networks to recognize images and sounds. Medical
professionals employ deep learning algorithms to detect and
diagnose brain cancers. Medical pictures, such as MRIs, can
be analyzed by these algorithms, and any areas of the brain
that might be harbouring a tumour can be pinpointed. It's
important to remember that deep learning can't replace the
knowledge of a radiologist or other medical professionals or
diagnostic tools. A brain tumour is a malignant cellular growth
that forms in the brain or skull. Migraines, seizures, trouble
speaking or seeing, and changes in mood or intelligence are
just some of the symptoms that may accompany a brain
tumour. Surgery, radiation therapy, and chemotherapy are all
viable choices for treating brain tumours, with the best course
of action being determined by the specifics of the tumour, its
location, and the patient's overall condition. A CNN could be
used to diagnose a brain tumour by first processing data from
diagnostic imaging tests like MRIs and CT scans. The CNN
would learn the tumor-indicating properties through training
on a labelled image collection. Following training, CNN may
examine new photos and provide predictions about the
presence of a cancer. Prediction precision is dependent on
both the training dataset and the CNN's architecture. This
work proposes using machine learning and deep learning to
discover and quantify brain tumors using MRI scans.
An optimized predictive model is generated using the
Fig. 3. Accuracy & Loss Analysis in ML &DL Approaches Comparison ensemble, a method that combines machine learning and deep
learning and provides a multitude of fundamental models.
C. Simulation Results of Brain Tumour Detection in ML & There have been numerous ensemble approaches documented
DL Approaches in the literature. Nevertheless, for the purposes of this study,
the method that received the greatest number of votes was
selected. Predominantly, the majority technique was
employed in categorization scenarios. A multitude of models
were utilized in order to generate recommendations for every
data point. All model predictions were considered. The bulk
of model forecasts determined final predictions. The proposed
method used Convolutional Neural Networks and SVM
ensemble models. This proposed research on brain tumor
prediction at various stages uses CNN's final fully-connected
layer properties nicely
V. CONCLUSION & DISCUSSION
We investigate the most problematic uses of machine
learning and deep learning for brain disease diagnosis in this
study. This survey provides unique insights on using ML and
Fig. 4. Tumor Detection in ML Classifier DL to diagnose and treat brain illnesses. The domains of

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.
 2024 Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE)
machine learning and deep learning ML & DL are becoming [5] Saladi, S., Karuna, Y., Koppu, S., Reddy, G. R., Mohan, S., Mallik, S.,
increasingly complex, making it more difficult to create & Qin, H. (2023). Segmentation and Analysis Emphasizing Neonatal
MRI Brain Images Using Machine Learning Techniques. Mathematics,
efficient methods of identification, feature extraction, and 11(2), 285.
classification. Researchers from all over the world are making [6] Doshi, R., Hiran, K. K., Prakash, B., & Vyas, A. K. (2023). Deep belief
a concerted effort to improve these procedures by exploring network-based image processing for local directional segmentation in
all available options. Improving the precision with which brain tumor detection. Journal of Electronic Imaging, 32(6), 062502.
classes are assigned is a key consideration. More training data [7] Panigrahi, A., & Subasi, A. (2023). Magnetic resonance imagining-
is needed to achieve this goal since more comprehensive based automated brain tumor detection using deep learning techniques.
training leads to more precise findings. Better outcomes may In Applications of Artificial Intelligence in Medical Imaging (pp. 75-
107). Academic Press.
be achieved by using hybrid algorithms that integrate
[8] ZainEldin, H., Gamel, S. A., El-Kenawy, E. S. M., Alharbi, A. H.,
supervised and unsupervised learning, as well as machine Khafaga, D. S., Ibrahim, A., & Talaat, F. M. (2022). Brain Tumor
learning and deep learning. After discussing several data Detection and Classification Using Deep Learning and Sine-Cosine
sources and feature extraction methods for brain diseases, it is Fitness Grey Wolf Optimization. Bioengineering, 10(1), 18.
obvious that the accuracy of the systems differs depending on [9] Kandimalla, S. Y., Vamsi, D. M., Bhavani, S., & VM, M. (2023).
the classifiers used and the feature extraction processes carried Recent Methods and Challenges in Brain Tumor Detection Using
out. To determine the limitations of current ML/DL-based Medical Image Processing. Recent Patents on Engineering, 17(5), 8-
23.
strategies for detecting various brain disorders, this study
involves a debate centred on a number of open research [10] Pan, Y.; Huang, W.; Lin, Z.; Zhu, W.; Zhou, J.; Wong, J.; Ding, Z.
Brain tumor grading based on neural networks and convolutional
problems. We observed that interoperability and the quality of neural networks. In Proceedings of the 2015 37th Annual International
training data are significant issues when creating ML and DL- Conference of the IEEE Engineering in Medicine and Biology Society
based solutions. The success of DL/ML algorithms hinges on (EMBC), Milan, Italy, 25–29 August 2015.
whether or not we can amass enough training data without [11] Vimal Kurup, R.; Sowmya, V.; Soman, K.P. Effect of Data Pre-
compromising their performance. More work needs to be done processing on Brain Tum r Classification Using Capsulenet. In
on a number of fronts, including resource efficiency, medical Proceedings of the International Conference on Intelligent Computing
and Communication Technologies, Hyderabad, India, 9–11 January
data administration on a massive scale, and security and 2019; Springer: Berlin/Heidelberg, Germany, 2020.
privacy, before ML/DL-based systems can be widely adopted. [12] Mukherkjee, D.; Saha, P.; Kaplun, D.; Sinitca, A.; Sarkar, R. Brain
Researchers interested in medical applications of artificial tumor image generation using an aggregation of GAN models with
intelligence and ML/DL for disease detection in the brain style transfer. Sci. Rep. 2022, 12, 9141.
should find this survey informative [13] Ari, A.; Hanbay, D. Deep learning based brain tumor classification and
detection system. Turk. J. Electr. Eng. Comput. Sci. 2018, 26, 2275–
As demonstrated in the antecedent research inquiry, deep 2286.
learning approaches yield significantly higher classification [14] Joshi, S.R.; Headley, D.B.; Ho, K.C.; Paré, D.; Nair, S.S. Classification
accuracies for brain MRIs than traditional ML techniques. of brainwaves using convolutional neural network. In Proceedings of
Conversely, deep learning algorithms exhibit a training the 2019 27th European Signal Processing Conference (EUSIPCO), A
requirement that is exponentially greater than that of Coruña, Spain, 2–6 September 2019; pp. 1–5.
conventional approaches. Recent research indicates that deep [15] Krishnammal, P.M.; Raja, S.S. Convolutional neural network based
image classification and detection of abnormalities in M.R.I. brain
learning methodologies have evidently surpassed the images. In Proceedings of the 2019 International Conference on
boundaries of intelligent and specialized systems and the Communication and Signal Processing (ICCSP), Melmaruvathur,
medical domain. Furthermore, there are limitations to the Tamil Nadu, India, 4–6 April 2019.
approaches that should be taken into account when diagnosing [16] Poonguzhali, N.; Rajendra, K.R.; Mageswari, T.; Pavithra, T.
and classifying brain tumors. As a result, our proposed Heterogeneous deep neural network or healthcare using metric
solution was compared to previous research in Table 1 in learning. In Proceedings of the 2019 IEEE International Conference on
System, Computation, Automation and Networking (ICSCAN),
terms of performance to demonstrate how integrating deep Pondicherry, India, 29–30 March 2019.
learning and machine learning models could result in [17] Raj, A.; Anil, A.; Deepa, P.L.; Aravind Sarma, H.; Naveen Chandran,
enhanced performance. The necessity for this study was R. BrainNET: A Deep Learning Network for Brain Tumor Detection
recognized through a review of prior research, which also and Classification. In Advances in Communication Systems and
advocated for improved performance to be attained via Networks; Springer: Singapore, 2020; pp. 577–589.
implementation. References and deep learning-based [18] Anilkumar, B.; Kumar, P.R. Tumor classification using block wise fine
approaches are encouraged in the methodologies presented in tuning and transfer learning of deep neural network and K.N.N.
classifier on M.R. brain images. Int. J. Emerg. Trends Eng. Res. 2020,
the literature, with CNN showing outstanding proficiency in 8, 574–583.
handling complicated problems
REFERENCES
[1] Siar, M., & Teshnehlab, M. (2019, October). Brain tumor detection
using deep neural network and machine learning algorithm. In 2019 9th
international conference on computer and knowledge engineering
(ICCKE) (pp. 363-368). IEEE.
[2] Komarasamy, G., & Archana, K. V. (2023). A novel deep learning-
based brain tumor detection using the Bagging ensemble with K-
nearest neighbor.
[3] Asad, R., Imran, A., Li, J., Almuhaimeed, A., & Alzahrani, A. (2023).
Computer-Aided Early Melanoma Brain-Tumor Detection Using
Deep-Learning Approach. Biomedicines, 11(1), 184.
[4] Ramtekkar, P. K., Pandey, A., & Pawar, M. K. (2023). Innovative brain
tumor detection using optimized deep learning techniques.
International Journal of System Assurance Engineering and
Management, 1-15.

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:04:54 UTC from IEEE Xplore. Restrictions apply.