Comparative Analysis of CNNs and Vision Transformer Models for

Brain Tumor Detection

Safa Jraba

, Mohamed Elleuch

, Hela Ltifi

and Monji Kherallah

National School of Electronics and Telecommunications (ENETCom), University of Sfax, Tunisia

National School of Computer Science (ENSI), University of Manouba, Tunisia

Faculty of Sciences and Techniques of Sidi Bouzid, University of Kairouan, Tunisia

Faculty of Sciences, University of Sfax, Tunisia

Keywords: Brain Tumor, Deep Learning, Diagnosis, EfficientNet, ViT, Medical Imaging, Classification, SVM.

Abstract: Brain tumors are irregular cell mixtures existing within the brain or central spinal canal. They could be

cancerous or benign. The likelihood of the best possible prognosis and therapy increases with the speed and

accuracy of detection. This work provides a method for detecting brain tumors that combines the capabilities

of vision transformers and CNNs. In contrast to other studies that primarily relied on standalone CNN or ViT

architectures, our method uniquely integrates these models with a Support Vector Machine classifier for the

improvement of accuracy and robustness in medical image classification. While the ViT makes it possible to

combine CNN and ViT to improve the accuracy of medical imaging of the disease, the CNN extracts

hierarchical features. In-depth analyses of benchmark datasets pertaining to imaging modalities and clinical

perspectives were conducted. According to the experimental findings, ViT and EfficientNet identified tumors

with an accuracy of 98%, while the greatest reported accuracy of 98.3% was obtained when ViT was

combined with an SVM classifier. Our findings suggest that our method may improve brain tumor detection

methods.

1 INTRODUCTION

The abnormal growth of cells inside the brain or

Central Spinal Canal is the first indication of a brain

tumor. They fall into two categories: malignant-

invasive and benign-non-invasive. Benign tumors are

less aggressive because they do not contain cancer

cells, and they tend to have very well-defined

periphery patterns. Generally, they are amenable to

surgical interventions. Even benign tumors can

potentially wreak havoc on a patient's health by

compressing sensitive sites or interfering with the

circulation of the cerebrospinal fluid. Whereas

malignant tumors are composed of cancer cells, their

ability to invade local tumors and easily spread to

adjacent tissues renders them potentially deadly.

Radiation therapy, chemotherapy, and surgical

intervention may be provided as treatment options.

https://orcid.org/0009-0007-7818-5091

https://orcid.org/0000-0003-4702-7692

https://orcid.org/0000-0003-3953-1135

https://orcid.org/0000-0002-4549-1005

Extremely large computational efforts are often

required by this process-altering image encoders into

pixel sequences for their scrutiny. In response,

Parmar et al. (2018) considered self-attention in local

patches only around the query pixels while avoiding

the cost of global computations over the full image.

Meanwhile, for medical imaging, especially in

brain tumor detection and identification, there is no

little tribulation, heavily influencing treatment plans

and patient outcomes. Recent developments in deep

learning have fueled potential unlocks in the

evaluation of medical images, though tumor

diagnosis-based medical image analysis remains to

date, still an area earning its mark in practical

applications. CNNs have shown promise in feature

extraction and classification in many applications,

including medical imaging. In parallel, Vision

Transformers (ViTs) appeared and became popular

1432

Jraba, S., Elleuch, M., Ltiﬁ, H. and Kherallah, M.

Comparative Analysis of CNNs and Vision Transformer Models for Brain Tumor Detection.

DOI: 10.5220/0013381900003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 1432-1439

ISBN: 978-989-758-737-5; ISSN: 2184-433X

because of their capability to extract distant

associations from image data, which affords a unique

interpretation methodology.

Medical imaging experts continue to have serious

challenges in the identification and diagnosis of brain

tumors, which accordingly has a profound impact on

treatment and patient outcomes. Deep learning

techniques have come recently with a whole different

ball game in medical image processing, allowing

pathways towards accuracy and efficiency in tumor

diagnosis that have never been encountered. These

provide effective tools for feature extraction and

classification in different medical imaging fields

using CNNs. Concurrently, Vision Transformers

(ViTs) are mushrooming into popularity for excelling

at extracting distant associations from image data;

hence, offering a potentially new abstraction of image

interpretation.

Especially, this paper aims to do complete and

thorough and full measure-in-dependent study on the

CNN and Vision Transformers for cerebral tumor

detection in medical images. Such systems regain

their respective advantages by using each of the

architectures' strengths for a more resilient tumor

detection system and more accurate local-based

statistics. With the proposed method, we intend to

utilize all possible local and global information in the

medical images concurrently to deliver more detailed

and accurate tumor structure analyses.

In this work, we carry out comprehensive studies

on standard brain tumor picture datasets, comparing

our hybrid approach's efficacy to both standalone

Vision Trans-former models and traditional CNN-

based techniques.

Our results show the potential for CNN

(EfficientNet B0) and Vision Transformer fusion

technique to improve the diagnostic accuracy and

simplify the clinical decision-making process, as

evidence by the usefulness of both methods in brain

tumor detection. We speculated about the effects this

work could have on future medical imaging research

and presented suggestions for improving techniques

for brain tumor detection.

Contributions

This paper makes the following contributions:

- Provide a hybrid approach for brain tumor

detection that combines EfficientNet and

Vision Transformers (ViT) with SVM.

- A thorough comparison of CNN, solo ViT,

and our suggested hybrid model.

- Reach cutting-edge outcomes on MRI

datasets with an accuracy of 98.3%.

2 RELATED WORKS

The search for precise detection methodologies for

cerebral tumors has sparked significant research

efforts in recent years. Numerous approaches have

been investigated to address this vital need in the field

of health. Conventional diagnostic modalities such as

magnetic resonance imaging (MRI) and computed

tomodensitometry (CT scans) have historically been

the primary tools used to identify brain tumors. Their

effectiveness in early detection and precise

delineation of terrorist borders, however, remains a

challenge.

To identify and comprehend these cerebral

tumors, specialized ondelettes and sup-port systems

have been used in conjunction with MRI. The precise

and automated classification of brain IRM pictures

holds significant value in medical research and

interpretation. Uncontrolled cell division that results

in aberrant cell clusters inside or outside the brain is

the cause of brain tumors. These aberrant cell clusters

harm healthy cells and interfere with regular brain

function. The objective was to distinguish between

brain tissue that was unaffected by tumors and brain

tissue that had tumors, whether benign or malignant.

Gurbină et al. (2019)

CNN applied to MRI images has been shown to

be beneficial in many recent studies for the

classification of brain-related illnesses. Yuan et al.

(2018).

Researchers and medical professionals can locate

the area of the brain afflicted by a tumor by using

MRI, an imaging method that shows the anatomy and

structure of the human brain. Sakhthidasan et al.

(2021)

Abd-Ellah et al. (2019) have presented an

enhanced method for detecting brain tumors in order

to identify malignant tumors. Due to the low contrast

of mous tissues, lesion detection is a challenging task

that requires the use of adaptive clustering k-means to

obtain a better segmentation method in order to

improve prediction accuracy.

A thorough review of well-known deep learning

models that are applied to various types of brain

tumor investigation by Waqas et al. (2020).

Amarapur et al. (2019) discussed both traditional

automatic learning methods and deep learning

techniques for the validation of cerebral tumors. With

the aid of deep learning approaches, they were able to

identify, segment, and classify brain tumors

effectively using three different algorithms, leading to

improved performance.

A high-level System Cancer Diagnosis by

coalescing the four Level-I taxonomy components

Comparative Analysis of CNNs and Vision Transformer Models for Brain Tumor Detection

1433

"DIV" Data, Image Segmentation processing and

VIEW is proposed by Laukamp et al. (2019). Level-I

taxonomy DIV evaluation consists of acceptance

Rate and Completion Rate by DL convoluiton neural

networks.

Archana et al. (2023) performed a comparative

study of several optimizers used in convolutional

neural networks (CNNs) to detect brain tumors from

medical photos. The study examined how well the

Adam and SGD optimizers performed when applied

with the AlexNet and Le-Net CNN architectures. The

study's findings, which were based on a dataset of

1547 images, showed that the AlexNet architecture

and Adam optimizer outperformed LeNet and SGD to

obtain an average accuracy of 94.76%.

The use of deep neural networks and machine

learning methods for the early diagnosis of brain

tumors using MRI data was investigated by Wani et

al. (2023). A variety of CNN architectures, including

AlexNet, GoogleNet, VGG-19, a bespoke model, and

a collection of machine learning models, were used in

the study. Significant emphasis was made on data

gathering, preprocessing, and classification

procedures in order to address class imbalance and

data heterogeneity. The group of models that attained

the maximum accuracy of 90.625% suggests that

combining deep and conventional machine learning

techniques has bright futures.

Rajinikanth et al. (2022) applied various pooling

techniques with the pre-trained VGG16 and VGG19

convolutional neural networks (CNNs) to identify

glioma and glioblastoma brain tumors from magnetic

resonance imaging (MRI) images. Employing the

ADAM Optimizer, the work has examined the

classification performances of the SoftMax, Decision

Tree (DT), k-Nearest Neighbors (KNN), and Support

Vector Machine (SVM) classifiers on 2000 images

acquired from The Cancer Imaging Archive (TCIA).

The results showed that DT based on average-pooling

and VGG16 managed to gain maximum classification

accuracy, which was 96.08%.

Recent developments in Vision Transformers

(ViT) have shown that they may effectively extract

features from images by utilizing self-attention

techniques, outperforming conventional

convolutional neural networks. Zhang et al. (2O22)

In this section, we look closely at a few recently

published, more successful techniques (See Table 1).

Table 1: Research on the detection of brain tumors.

Authors Techni

ues Dataset Accurac

Archana et

al. (2023)

CNN,

AlexNet,

LeNet

1547

images of

Brain

Tumor

Dataset

94.76%.

Wani et al.

(2023)

AlexNet,

GoogleNet,

VGG-19,

Brain tumor

Dataset

MRI scans.

90.625%

Rajinikanth

et al. (2022)

VGG16

VGG19

SVM, KNN

MRI

images

96.08%

Laukamp et

al. (2019)

Deep

learning

Model

Data View

(DIV)

95,08%

Our study produced impressive results that

outperformed previous methods. Through the

automation and optimization of the diagnostic

procedure, this research has a significant potential to

improve early brain tumor detection, which could

improve patient care and quality of life. Our findings

advance the state of the art in the categorization of

cerebral tumors and demonstrate the effectiveness of

transfer learning in the analysis of medical images.

3 METHODOLOGIES

This section describes the process of developing our

experiments to improve the accuracy of brain tumor

detection. The first step is to collect data. The second

stage involves data pre-processing, such as data

augmentation techniques. The last step is to predict

the outcome, where the ViT and EfficientNet B0

models are applied.

3.1 Gathering Data

The tomodensitometry (CT) and magnetic resonance

imaging (MRI) are two of the many biomedical

imaging techniques that are vital for the detection of

brain tumors. The MRI, which is especially well-

known for its high resolution and detailed

information, provides better viewpoints. Table 2

displays the MRI dataset from the well-known

Kaggle website, which had 3264 images categorized

into three groups.

Table 2 provides a clear overview of the training

and test data distribution across classes, now

including percentages to highlight data balance or

imbalance for glioma, meningioma, and pituitary

categories.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1434

Table 2: Contents of a collection for brain MRI scans.

Data Label

Class Train set (%) Test set (%)

Glioma 40% 42%

Meningioma 35% 33%

Pituitar

25% 25%

3.2 Pre-Processing

Our brain tumor MRI dataset underwent a number of

pre-processing procedures to guarantee data

consistency and quality. In order to improve model

understanding and prediction accuracy, these steps

included resizing images to a standard dimension,

normalizing the data to improve model convergence,

and using augmentation techniques to diversify the

dataset and represent different tumor classes, such as

glioma, meningioma, pituitary (See Figure 1).

Figure 1: Using the MRI brain tumor data augmentation

technique.

Table 3: Augmentation techniques used.

Techni

ue Parameters Ob

ective

Rotation [-30°, +30°] Introduce angle

variations

Translation [-0.2, +0.2] Simulate

horizontal/vertical

shiftin

Scaling [0.8, 1.2] Represent

different tumor

sizes

Horizontal/vertic

al fli

Random Reduce

directional bias

Luminosity/

Contrast

[0.5, 1.5] Simulate different

lighting

conditions

The data augmentation methods used to increase

model generalization and resilience in brain tumor

detection tasks are shown in Table 3.

3.3 Proposed Architectures

3.3.1 Vision Transformer (ViT)

Vision Transformer model debuted at the 2021

International Conference on Learning Representations

(ICLR) in the paper "An Image is Worth 16x16 Words:

Transformers for Image Recognition at Scale." ViTs

adapt transformer topologies from natural language

processing (NLP) to convert input images into patches

resembling word tokens.

Figure 2: Vision Transformers (ViT) architecture.

Brain tumor images are converted into "patches"

for examination. The raw image is divided into small

sections prior to applying the "patches," which are

then processed by the ViT model to produce

insightful representations. This transformation

process, ViT can effectively identify and categorize

aspects of brain tumors as well as other relevant

components in the medical image.

For classification, the standard ViT employs a

Feed-Forward Network (FFN) (See Figure 2). In our

approach, we replaced the FFN classifier with a

Support Vector Machine (SVM) to leverage its proven

performance in the medical field. Yang et al. (2018)

The Support Vector Machine (SVM) classifier is

utilized for brain cancer detection due to its ability to

handle high-dimensional data and create a robust

decision boundary between different classes (See

Figure 3). Rajinikanth et al. (2022)

By finding the optimal hyperplane that maximizes

the margin be-tween tumor and non-tumor samples,

the SVM can accurately classify new, unseen images

as either containing a brain tumor or not. Its

effectiveness in medical applications, including

cancer detection, stems from its ability to minimize

classification errors and enhance predictive accuracy.

Garcia et al. (2020)

Comparative Analysis of CNNs and Vision Transformer Models for Brain Tumor Detection

1435

Figure 3: SVM approaches; (a) one-versus-all method, (b)

one-versus-one method. Tan et al. (2023).

3.3.2 EfficientNet B0

A convolutional neural network (CNN) with limited

processing and memory capacity, EfficientNet B0 is

intended for deep learning applications. Due to the

simultaneous optimization of the CNN models' depth,

width, and resolution, it has a lightweight but robust

architecture (See Figure 4).

Figure 4: Efficient Net B0 architecture.

With the release of EfficientNetV2, model

performance and training speed were enhanced,

making it especially appropriate for computationally

demanding jobs like medical imaging. Zhang et al.

(2023)

4 EXPERIMENTS AND RESULTS

This section describes the data collection process; the

images of cerebral tumors serve as the basis for the

data used in this work. Effective categorization

results are achieved when a broad and diverse data set

is used. We will describe and discuss the

experimental configurations in the following. The

obtained results are then presented and contrasted

with the suggested systems.

We investigate the potent capabilities of CNN and

Vision Transformers (ViT) architectures to solve the

challenging task of picture classification on the

cerebral tumor data game. Conventional

convolutional neural networks (CNNs) have been the

architecture of choice for image-related tasks, but

vision-related tasks (ViTs) introduce auto-attention

mechanisms inspired by the Transformer

architecture, which was originally designed for

natural language processing.

Furthermore, in order to actually carry out the

study, a phase of analysis and discussion of the

experiment's parameters is required (See Table 4).

Table 4: Parameters of EfficientNet B0 Model.

Total parameters 4,054,695

Trainable

arameters 4,012,672

Non-trainable

arameters 42,023

In this work, we combined a Support Vector

Machine (SVM) for classification with Vision

Transformers (ViT) as the foundation for feature

extraction. The following are specifics of the SVM

configuration that was used:

 Kernel Function: Because radial basis

function (RBF) can handle nonlinear feature

spaces, that's the kernel we used.

Kernel Parameters for RBF:

 Sigma: The value of sigma, also called

gamma, is fixed at 0.01. The influence of a

single training point is controlled by this

parameter.

 Coefficient: A second parameter was set to

0.5. For some RBF kernel implementations, it

is commonly referred to as coef0. The impact

of the linear term in the polynomial and

sigmoid kernels is adjusted by this coefficient.

4.1 Dataset

An accurate dataset is essential for using CNNs to

classify brain tumors into Glioma, Meningioma, and

Pituitary categories. Our dataset, sourced from

medical facilities, includes three distinct labels for

tumor classification:

 Glioma: classification for brain tumors with

gliomas as its defining feature.

 Meningioma: The categorization of brain

tumors classified as meningiomas

 Pituitary: Brain tumors associated with the

pituitary gland.

The gathered dataset is split into two sub-sets: the

first represents the 80% training portion and the

second, the 20% test portion. There are three classes

of images for each of the two image portions, and

each class corresponds to a distinct label for brain

tumor classification (See Figure 5).

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1436

Figure 5: Samples of brain tumor labels.

4.2 Results and Discussion

One of the most widely used systems, the

Convolutional Neural Network (CNN) model, is used

to classify brain tumors into the Glioma, Meningioma,

and Pituitary classes. The performance of the CNN

convolutional neural network, the EfficientNet B0, and

the Vision Transformers (ViT) architectures are the

main subjects of the experimental investigation. The

following figures (6 and 7) provide a summary of our

suggested system's performance.

4.2.1 Vision Transformers ViT

Figure 6: Sample of Brain tumor image applying ViT.

Figure 6 illustrates a brain tumor MRI image

processed by the Vision Transformer (ViT), dividing

the image into grid-like patches for feature extraction

and classification.

Figure 7: Performance of ViT approach.

The standard ViT shows equable scores of 98% in

measures of precision, recall, F1 score, and accuracy,

demonstrating reliable classification. Using ViT and

SVM, we get an enhanced precision of 98.39% and an

accuracy of 98.3%; this means reduced false positives,

while recall drops slightly to 98.03%. The F1 slightly

increases to 98.2% signifying an overall improvement.

These metrics indicate that ViT can achieve very

good brain tumor detection by accurately

distinguishing tumor images from the rest of the

medical images. (See Table 5)

Table 5: Classification report of our proposed method (ViT).

Methods/

Metrics

Precision Recall

F1-

Score

Accuracy

Standard ViT 98% 98% 98% 98%

ViT with

SVM

98,39% 98,03% 98,2% 98,3%

Figure 8: Confusion matrices using ViT.

Figure 9: Confusion matrices using ViT combined with

SVM.

Figures 8 and 9 presents the confusion matrices

generated for the image brain tumor detection task

using ViT and ViT combined with SVM.

Comparative Analysis of CNNs and Vision Transformer Models for Brain Tumor Detection

1437

Figure 10: Confusion matrix of the proposed methods

(Efficient Net B0).

For the Standard ViT model, with perfect balance

between them. It predicts the majority of cases

correctly, and only makes minor errors, especially in

the Gliomas class because of its small number of false

negatives. On the whole, the model is performing

well, with only the slightest hit to certain classes.

All three metrics indicate extremely high

precision and recall (98%).

On the other hand, the ViT with SVM slightly

improves overall precision (98.39%) against the

Standard ViT according to incorrect predictions of a

small number of false negatives identified with the

Glioma class. In some classes, the performance on

Meningioma was a little compromised; however, this

model establishes a more balanced mechanism across

the classes, with an overall increase due to lesser

errors expected, thus proving this is an efficient and a

more reliable model to perform tumor detection.

4.2.2 Efficient Net B0

To evaluate a brain tumor detection model's efficacy,

its ability to classify gliomas, meningiomas, and

pituitary tumors is assessed. EfficientNet B0's

confusion matrix (Figure 9) shows true positives

(TP), true negatives (TN), false positives (FP), and

false negatives (FN). Metrics like precision, recall,

and specificity are computed to measure its

performance.

Plotting the metrics for the 98% accurate Efficient

Net B0 architecture would undoubtedly shed light on

the model's performance throughout evaluation and

training. Loss and additional measures like validation

accuracy, validation loss, and confusion matrix are

frequently presented alongside accuracy (See Figure

11).

Figure 11: Accuracy vs Epoch and Loss vs Epoch Graph

(Efficient Net B0).

Table 6: Classification report of our proposed methods

(Efficient Net B0).

Precision Recall

F1-

score

Support

Glioma

0.98 0.96 0.97 93

Meningioma

0.98 1.0 0.98 51

Pituitary

0.98 0.98 0.98 96

Accuracy

0.98 327

Macro avg

0.98 0.98 0.98 327

Weighted avg

0.98 0.98 0.98 327

The effectiveness of a brain tumor detection

model is assessed in this work. Accompanying the

support values for every class are f1-score, recall, and

precision metrics. The model's promise for brain

tumor diagnosis is demonstrated by its high precision,

recall, for all classes, as well as its overall accuracy

of 98% (see Table 6). The CNN and ViT hybrid

strategy make use of complementary strengths.

Likewise, SVMs continue to be reliable classifiers for

tasks involving medical imaging. He et al. (2022)

Table 7: Performance Comparison using Brain tumor

Dataset.

Authors Architecture Accuracy

Our Proposed

Vision Transformer

ViT

ViT with SVM

EfficientNet B0

98%

98.3%

98%

Wani et al.

(

2023

)

CNN, AlexNet, LeNet

94.76%

Rajinikanth et

al.

(

2022

)

AlexNet, GoogleNet,

VGG-19

90.62%

Zhang et al.

(

2022

)

VGG16, VGG19,

SVM, KNN

96.08%

Pang et al.

(

2023

)

Deep Learning Model 95.08%

Table 7 compares various brain tumor detection

methods. Our models (EfficientNet, ViT, and SVM)

achieved a maximum accuracy of 98.3%,

outperforming previous methods like VGG19,

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1438

AlexNet, GoogleNet, and KNN. This demonstrates

the effectiveness of our lightweight and efficient

architectures, such as Net B0 and ViT, in achieving

promising results for early tumor diagnosis with high

accuracy.

4 CONCLUSIONS

Our investigation into EfficientNet and Vision

Transformer methods yielded a notable 98.3%

accuracy for brain tumor identification. By

combining Vision Transformer, SVM, and

EfficientNet, we developed a reliable MRI-based

detection system. Future work could explore Faster

R-CNN integration to improve the speed and

accuracy of tumor diagnosis, potentially enhancing

patient outcomes and advancing brain tumor

detection.

REFERENCES

N. Parmar, A. Vaswani, J. Uszkoreit, L. Kaiser, N. Shazeer,

A. Ku, and D. Tran. (2018).Image transformer,

International conference on machine learning. PMLR,

pp. 4055–4064

M. Gurbină, M. Lascu and D. Lascu. ( 2019 ).Tumor

Detection and Classification of MRI Brain Image using

Different Wavelet Transforms and Support Vector

Machines, 42nd International Conference on

Telecommunications and Signal Processing (TSP).

Yuan, Lin, et al.(2018).multi-center brain imaging

classification using a novel 3D CNN approach, IEEE

Access 6 , 49925-49934

K.Sakthidasan, A S Poyyamozhi, Shaik Siddiq Ali, and

Y.Jennifer. ( 2021).Automated Brain Tumor Detection

Model Using Modified Intrinsic Extrema Pattern based

Machine Learning Classifier Fourth International

Conference on Electrical, Computer and

Communication Technologies.

M. K. Abd-Ellah, A. I. Awad, A. A. Khalaf, and H. F.

Hamed. (2019). A review on brain tumor diagnosis

from MRI images: Practical implications, key

achievements, and lessons learned, Magnetic resonance

imaging, vol. 61, pp. 300-318.

M. Waqas Nadeem, Mohammed A. Al Ghamdi, Muzammil

Hussain, Muhammad Adnan Khan, Khalid Masood

Khan, Sultan H. Almotiri and Suhail Ashfaq

Butt.(2020). Brain Tumor Analysis Empowered with

Deep Learning: A Review, Taxonomy, and Future

Challenges Brain Sci.

B. Amarapur.(2019).Cognition-based MRI brain tumor

segmentation technique using modified level set

method, Cognition, Technology & Work, vol. 21, pp.

357-369.

K. R. Laukamp, F. Thiele, G. Shakirin, D. Zopfs, A.

Faymonville, M. Timmer, et al.(2019). Fully automated

detection and segmentation of meningiomas using deep

learning on routine multiparametric MRI, European

radiology, vol. 29, pp. 124-132.

B. Archana, M. Karthigha, and D. Suresh Lavanya. (2023).

Comparative Analysis of Optimisers Used in CNN for

Brain Tumor Detection, Institute of Electrical and

Electronics Engineers (IEEE).

S. Wani, S. Ahuja, and A. Kumar.(2023).Application of

Deep Neural Networks and Machine Learning

algorithms for diagnosis of Brain tumour, in

Proceedings of International Conference on

Computational Intelligence and Sustainable

Engineering Solution, Institute of Electrical and

Electronics Engineers Inc. pp. 106–111.

V. Rajinikanth, S. Kadry, R. Damasevicius, R. A. Sujitha,

G. Balaji, and M. A. Mohammed. (2022).Detection in

Brain MRI using Pre-trained Deep-Learning Scheme,

the 3rd International Conference on Intelligent

Computing, Instrumentation and Control Technologies:

Computational Intelligence for Smart Systems,

ICICICT

He, K., Zhang, X., Ren, S., & Sun, J. (2022). Synergizing

Convolutional and Transformer Models for Vision

Tasks. IEEE Transactions on Pattern Analysis and

Machine Intelligence, 44(8), 4257-4268.

Yang, Z. R. (2018). Biological applications of support

vector machines. Briefings in bioinformatics, 5(4), 328-

338.

Jraba, S., Elleuch, M., Ltifi, H., & Kherallah, M. (2023).

Alzheimer Disease Classification Using Deep CNN

Methods Based on Transfer Learning and Data

Augmentation. International Journal of Computer

Information Systems and Industrial Management

Applications, 1-10..

Cervantes, J., Garcia-Lamont, F., Rodríguez-Mazahua, L.,

Lopez, A. (2020). A comprehensive survey on support

vector machine classification: Applications, challenges

and trends. Neurocomputing, 408, 189-2

Tan, M., Pang, R., & Le, Q. V. (2023). EfficientNetV2:

Smaller models and faster training. International

Conference on Machine Learning (ICML), 10023-

10033.

Comparative Analysis of CNNs and Vision Transformer Models for Brain Tumor Detection

1439