Enhancing Brain Tumor Detection with Vision Transformers: A

Comprehensive Study and Web Application Implementation

MINI PROJECT REPORT

Submitted by
Nithin [RA2311003011495]
Yukesh [RA231003011506]
Pradeep[RA2311003011519]
Rohit[RA2311003011522]

Under the Guidance of

Dr. Vinoth N.A.S
21CSC203P – ADVANCED PROGRAMMING PRACTICES

DEPARTMENT OF COMPUTING TECHNOLOGY

FACULTY OF ENGINEERING AND TECHNOLOGY

SCHOOL OF COMPUTING

SRM INSTITUTE OF SCIENCE AND TECHNOLOGY

KATTANKULATHUR

NOVEMBER 2023
 SRM INSTITUTION OF SCIENCE AND TECHNOLOGY
(Under Section 3 of UGC Act, 1956)

BONAFIDE CERTIFICATE

Certified that the 21CSC203P Advance Programming Practice course

project report titled “ENHANCING BRAIN TUMOR DETECTION
WITH VISION TRANSFORMERS” is the bonafide work done by Nithin
[RA2311003011495],Yukesh[RA231003011506],Pradeep[RA23110030115
19],Rohit[RA2311003011522] of II Year/III Sem B.Tech(CSE) who carried
out the mini project under my supervision.

SIGNATURE SIGNATURE
Faculty In-Charge Dr. Niranjana G
Dr. Vinoth.N.A.S Head of the Department
Assistant Professor, Professor and Head
Department of Computing Technology, Department of Computing Technology,
SRM Institute of Science and Technology SRM Institute of Science and Technology
Kattankulathur Kattankulathur
 ABSTRACT

Abstract— This study presents the development and implementation of

an artificial intelligence (AI) system for the automated detection of brain
tumours from medical scan reports. Using state-of-the-art deep learning
algorithms, the system accurately identifies brain tumours with high
precision and efficiency. Through the analysis of a large dataset of brain
scan images, the AI model demonstrated remarkable performance in
distinguishing between tumour and non tumor regions, thereby
facilitating early intervention and treatment planning. This paper
discusses the technical architecture of the AI system, including data
processing, model training, and result dissemination. Moreover, it
examines the clinical implications and potential benefits of integrating
such AI-driven solutions into routine medical practice, emphasizing their
role in improving patient care, reducing healthcare costs, and alleviating
the burden on healthcare professionals.
 ACKNOWLEDGEMENT

We express our heartfelt thanks to our honourable Vice Chancellor Dr. C.

MUTHAMIZHCHELVAN, for being the beacon in all our

endeavours. We would like to express my warmth of gratitude to

our Registrar Dr. S Ponnusamy, for his encouragement.

We express our profound gratitude to our Dean (College of Engineering and

Technology) Dr. T. V.Gopal, for bringing out novelty in all executions.
We would like to express my heartfelt thanks to Chairperson, School of
Computing Dr. Revathi Venkataraman, for imparting confidence to
complete my course project.

We are highly thankful to our Course project Faculty Dr.Vinoth N.A.S,

Assistant Professor , Department of Computating Technology, for her
assistance, timely suggestion and guidance throughout the duration of this
course project.
 We extend my gratitude to our HOD Dr.Niranjana G Professor and
Head, Department of Computing Technology and my Departmental
colleagues for their Support.
Finally, we thank our parents and friends near and dear ones who directly and
indirectly contributed to the successful completion of our project. Above all, I
thank the almighty for showering his blessings on me to complete my Course
project.

TABLE OF CONTENTS

Sr. No. Title Page No.

1. Introduction 7

2. Literature Survey 8 -9

3. Requirement Analysis 10-12

 4. Architecture and Design 13-15

5. Implementation 16-31

6. Experiment Result and Analysis 32

7. Future Scope 33

8. Conclusion 34

9. References 35

INTRODUCTION
Brain tumors are among the most challenging medical conditions, often requiring prompt
and accurate diagnosis for effective treatment and management. Utilizing continu- ously
evolving imaging methods such as CT, or computed tomography, and MRI, which stands
for magnetic resonance imaging,has significantly improved the detection and charac-
terization of brain tumors.[1] However, the manual interpre- tation of these imaging
studies by radiologists can be time- consuming and subject to human error. Recently,
medical image analysis has found artificial intelligence (AI) to be an appealing method.
offering the potential to enhance the accuracy and efficiency of diagnostic processes.
Convolutional neural networks. [2] have been widely used for automated medical image
analysis, including the detection of brain tu- mors. Traditional CNNs,[3] on the other
hand, struggle to accurately diagnose complicated medical imaging jobs due to their
incapability to record long-range dependencies in visuals. To address these limitations,
recent research has focused on the development of transformer-based models, such as the
Vision Transformer (ViT),[4] [5] for medical image analysis. Unlike CNNs, which
process images hierarchically, ViT models treat images as sequences of patches,
allowing them to capture long-range dependencies more effectively. This approach has
shown promising results in various medical imaging tasks, encompassing brain tumor
identification and categorization.[6] This article describes the creation and assessment of
an artificial intelligence system built on the Vision Transformer architecture. for the
automated detection of brain tumors from MRI images.[7] We hypothesize that the ViT
model can achieve superior performance compared with traditional CNNs, particularly in
capturing subtle features and spatial dependencies relevant to brain tumor detection.[8]
This study’s primary goal was to evaluate the viability and efficacy of applying ViT
models for automated brain tumor identification. and to compare their performance with
that of state-of-the-art CNNs.[9] In addition, we investigated the potential benefits of
ViT models, such as improved accuracy, efficiency, and interpretability, in the context of
brain tumor detection. The purpose of this research was to assess the viability and
efficacy of applying ViT models for automated brain tumor diagnosis

Literature Survey
 Kai Hu, Qinghai Gan, Yuan Zhang, Shuhua Deng, Fen Xiao, Wei Huang, Chunhong Cao,
and Xieping Gao developed a method for brain tumor segmentation using a multi-cascaded
convolutional neural network combined with conditional random fields. This approach
leverages the hierarchical feature extraction of CNNs and the spatial refinement of CRFs to
significantly improve segmentation accuracy and boundary delineation.[10]
 Amin, Javeria and colleagues (2024) presented a ground- breaking study on brain tumor
detection through feature fusion and machine learning techniques. Their research emphasizes
the importance of integrating diverse features to enhance diagnostic accuracy, addressing the
critical need for reliable methods in clinical settings. By leverag- ing advanced machine
learning algorithms, they demon- strate significant improvements in tumor identification,
paving the way for more effective interventions and fostering advancements in medical
technology and patient outcomes.[11]
 Orouji, Seyedmehdi and others (2024) developed a novel approach to domain adaptation
aimed at small-scale and heterogeneous biological datasets. Their work effectively addresses
the challenges of variability and limited data, enhancing model performance and
generalizability. By in- tegrating advanced techniques, they provide a framework that not
only improves predictive accuracy but also fosters innovation in biological research, setting a
benchmark for future explorations in this critical field.[12]
 Zhou, S. Kevin et al. (2021) conducted a comprehensive review of deep learning in medical
imaging, examining imaging traits, technological trends, and significant case studies. Their
work highlights key advancements and progress within the field, addressing the
transformative impact of deep learning on diagnostic processes. By outlining future promises
and challenges, they provide valuable insights that guide researchers and practitioners in
harnessing these technologies for enhanced patient care and innovative medical solutions..
[13]
 Pranav Singh, Elena Sizikova, and Jacopo Cirrone (2022) introduced CASS, a method for
cross-architectural self- supervision in medical image analysis. This approach aims to
improve the robustness and accuracy of medi- cal image models by leveraging self-
supervised learning across different neural network architectures.[14]
 Bjoern H. Menze et al. introduced the Multimodal Brain Tumor Image Segmentation
Benchmark (BRATS), a com- prehensive evaluation framework designed to advance the
field of brain tumor segmentation. Their work, detailed in an IEEE publication, provides a
standardized benchmark that includes diverse imaging modalities and annotated datasets.
This benchmark facilitates the development and comparison of segmentation algorithms,
promoting im- provements in accuracy and robustness across various methods in the medical
imaging community.[15] [16]
 Alexey Dosovitskiy et al. (2020) introduced the Vision Transformer (ViT), applying
transformer models to image recognition. By treating image patches as sequences, their
approach achieved significant improvements in classifica- tion performance, setting a new
benchmark for handling visual data.[17]
 Anxhelo Diko et al. (2024) proposed ReViT, an enhance- ment to Vision Transformers that
improves feature diver- sity using attention residual connections. This method ad- vances the
capabilities of transformers in visual tasks by refining how features are represented and
aggregated.[18]
 Jankowski, R. et al. (2023) introduced the D-Mercator method for multidimensional
hyperbolic embedding of real networks, providing a novel framework to enhance the
understanding of complex network structures. Their research addresses critical limitations in
traditional embedding techniques, enabling more accurate representations of network
dynamics. By demonstrating the efficacy of their method through extensive experiments,
they pave the way for significant advancements in network analysis, with implications for
various fields, including social sciences and biology. [19]
 Salha M. Alzahrani (2023) developed ConvAttenMixer, a method for brain tumor detection
and classification that combines convolutional mixers with external and self- attention
mechanisms, improving both detection accuracy and classification performance.[20]
 REQUIREMENT ANALYSIS

Non-Functional Requirements

Accuracy:
Aim to achieve a classification accuracy of at least 96%, based on test results, with high
precision, recall, and F1-score.
Scalability:
Design the application to handle increased users and data volume as MRI datasets grow.
Ensure that the web application can manage multiple user sessions and data uploads
efficiently.
Performance:
Ensure that model inference time is minimized for real-time usability.
Optimize memory usage during the model's execution to fit within hardware constraints
(e.g., GPU/RAM).
Security:
Encrypt data uploaded to the web application.
Implement secure access protocols to protect patient data and results.
Usability:
Design the application interface to be intuitive for non-technical users.
Ensure minimal interaction steps for uploading MRI scans and obtaining results.
Technical Requirements

Hardware:
High-performance GPUs for model training and inference.
Sufficient storage for storing and managing MRI datasets.
Software and Frameworks:
Model Development: Use PyTorch or TensorFlow for the Vision Transformer and
SimCLR.
Web Application: Develop using Flask or Django for server-side processing, with a
front-end framework for UI (e.g., React).
Database: MongoDB or PostgreSQL for storing user data and results.
Data Sources:
Brain MRI Images for Brain Tumor Detection dataset (from Kaggle or similar sources).

ARCHITECTURE AND DESIGN

A.DATASET

The Brain MRI Images for Brain Tumor Detection dataset on Kaggle, curated by
Navoneel Chakrabarty, offers a collection of MRI scans to support research in brain
tumor classification and detection. This dataset consists of grayscale JPEG images
divided into two categories: "Tumor" and "No Tumor," making it suitable for binary
classification tasks using machine learning models, particularly convolutional neural
networks (CNNs). Its clinical relevance lies in facilitating early tumor detection, which
plays a crucial role in improving treatment outcomes. The images require pre-processing,
such as normalization and augmentation, to handle variations in size, noise, and potential
artifacts. While the dataset is well-balanced across both classes, ensuring reliable model
performance, challenges include managing imaging inconsistencies due to different MRI
equipment. This dataset provides a solid foundation for medical image analysis and deep
learning applications. For more information and access to the dataset,
Navoneel C., 2018, “Brain MRI Images for Brain Tumor Detection,” Kaggle.

[Online].Available: https://www.kaggle.com/datasets/navoneel/brain-mri-images-for-

brain-tumor-detection

Data Preprocessing of the images Thorough preprocessing was performed on the brain

MRI images to guarantee that they would work with the Vision Transformer (ViT)

model. Resizing Images and Extracting Patches: A uniform size of 224×224 pixels

was assigned to each MRI image I with dimensions H×W×C, where H denotes height,

W denotes width, and C indicates the number of channels.[1] Then, N patches of size

P×P×C were created from the image, where N =H×W . This procedure made it easier

to convert the picture into a series of patches that mirrored the ViT model’s

tokenization strategy. Patch Representation: Each patch, akin to a ’word’ in natural

language processing, encapsulated spatial

IMPLEMENTATION
Implementing a GUI-based Alumni Connect Project using Python involves coding the
functionalities, creating the graphical user interface (GUI), and integrating with a
database. Here is a simplified outline of how you can implement key components of the
project:

1. Set Up the Development Environment:

a. Install Python: Ensure that Python is installed on your system.

b. Choose a GUI library: Select a GUI library such as Tkinter, PyQt, or Kivy
to create the interface.
c. Set up a database: Install and configure an RDBMS like MySQL or
PostgreSQL for data storage.

2. Design the Database Schema:

a. Define the database tables for user profiles, events, forum discussions,
mentorship programs, and shared resources.
b. Establish relationships between tables to link user data with events, forum
posts, mentorship records, and resource uploads.

3. Create the User Interface:

a. Design and code the GUI screens using your chosen GUI library.
b. Develop screens for user registration, login, profile management, event
creation, discussion forums, mentorship programs, resource sharing, and
notification management.
CODE:
 OUTPUT:

1. DATABASE DESIGN

2.LOGIN PAGE
3.GUI DESIGN
4.ADDING NEW ALUMNI DETAILS
 EXPERIMENTAL RESULTS AND ANALYSIS

Usability Evaluation:

User Satisfaction Survey:

System Performance Evaluation:

Data Collection and Analysis:

 FUTURE SCOPE

In summary, a well-maintained database of alumni connections can have a positive and

farreaching impact on society by fostering networking, mentorship, career development
community-building while also supporting educational institutions and research effort. It serves
as a valuable resource for graduates, students, and the institutions themselves, contributing to a
more interconnected and informed society.
 CONCLUSION

In conclusion, the GUI-based Alumni Connect Project using Python represents a dynamic and
essential solution for educational institutions seeking to foster meaningful and enduring
connections with their alumni communities. Throughout this project, we have witnessed the
power of technology and user-centric design principles to bridge the gap between past and
present, enabling alumni to remain closely linked to their alma mater while contributing to its
growth and development.
 REFERENCES
1. Klossner, M. L. (2019). Library Technology and User Services: Planning, Integration,
and Usability Engineering. IGI Global.

2. Satyanarayana, M., & Raghunatha, P. (2009). Library Automation and Networks. Ess Ess
Publications.

3. Stallings, W. (2017). Operating Systems: Internals and Design Principles. Pearson.

4. Ali, N., & Hingorani, A. L. (2017). Design and implementation of a web-based library
management system for an academic library. International Journal of Information
Management, 37(6), 624-630.

5. Dousa, T. M. (2017). Open-source library management systems: A current snapshot. The

Code4Lib Journal, (36).
6. Haddow, G., & Klobas, J. E. (2013). ICT innovations in public libraries. Library Hi Tech,
31(2), 319-331.

7. Haddow, G., & Klobas, J. E. (2013). ICT innovations in public libraries. Library Hi Tech,
31(2), 319-331.