Software Requirements

Specification

for

Brain Tumor Classification using

Vision Transformer

Prepared by

Group Number: 16

Name Roll No. Email-ID Signature

Shruti Patil 2103048 2103048@ritindia.edu
Atharva Kabade 2103025 2103025@ritindia.edu
Om Jadhav 2103034 2103034@ritindia.edu
Khushi Deshmukh 2103045 2103045@ritindia.edu

Course Name: Capstone Project

Project Guide: Prof. T. S. Ruprah

Head of Dept: Dr. S.U. Mane

Date: 23 / 07 / 2024
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page ii

Contents
REVISIONS ...................................................................................... ERROR! BOOKMARK NOT DEFINED.
1 INTRODUCTION .................................................................................................................................. 1
1.1 DOCUMENT PURPOSE .................................................................................................................... 1
1.2 PRODUCT SCOPE ........................................................................................................................... 1
1.3 INTENDED AUDIENCE AND DOCUMENT OVERVIEW ............................................................................ 1
1.4 DEFINITIONS, ACRONYMS AND ABBREVIATIONS................................................................................ 2
1.5 DOCUMENT CONVENTIONS ............................................................................................................. 2
1.6 REFERENCES AND ACKNOWLEDGMENTS ......................................................................................... 2
2 OVERALL DESCRIPTION ................................................................................................................... 3
2.1 PRODUCT PERSPECTIVE ................................................................................................................. 3
2.2 PRODUCT FUNCTIONALITY .............................................................................................................. 3
2.3 USERS AND CHARACTERISTICS ....................................................................................................... 4
2.4 OPERATING ENVIRONMENT ............................................................................................................. 4
2.5 DESIGN AND IMPLEMENTATION CONSTRAINTS.................................................................................. 5
2.6 USER DOCUMENTATION.................................................................................................................. 5
2.7 ASSUMPTIONS AND DEPENDENCIES ................................................................................................ 6
3 SPECIFIC REQUIREMENTS ............................................................................................................... 7
3.1 EXTERNAL INTERFACE REQUIREMENTS ........................................................................................... 7
3.2 FUNCTIONAL REQUIREMENTS ......................................................................................................... 8
3.3 BEHAVIOUR REQUIREMENTS ........................................................................................................... 9
4 OTHER NON-FUNCTIONAL REQUIREMENTS ............................................................................... 10
4.1 PERFORMANCE REQUIREMENTS ................................................................................................... 10
4.2 SAFETY AND SECURITY REQUIREMENTS ........................................................................................ 11
4.3 SOFTWARE QUALITY ATTRIBUTES ................................................................................................. 12
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 1

1 Introduction
1.1 Document Purpose
The primary purpose of this document is to provide a comprehensive overview of the "Brain Tumor
Detection using Vision Transformer" project. This document outlines the project's objectives,
technical specifications, and operational procedures involved in utilizing Vision Transformer (ViT)
technology for the accurate detection and classification of brain tumors. The system is designed to
enhance diagnostic accuracy and efficiency in medical settings by leveraging ViT's advanced
capabilities to analyze medical imaging data and assist healthcare professionals in making informed
decisions about patient care and treatment strategies.

1.2 Product Scope

The purpose of this project is to develop a deep learning-based system for accurate brain tumor
detection in noisy medical images. This system aims to address the limitations of traditional methods
by offering faster processing times and potentially improved diagnostic accuracy, even in the
presence of image noise. By achieving high accuracy in brain tumor detection, this project has the
potential to significantly impact patient care. Faster and more accurate diagnoses can lead to earlier
intervention, improved treatment outcomes, and ultimately, better patient prognoses.

1.3 Definitions, Acronyms and Abbreviations

 Deep Learning (DL): A subfield of machine learning using artificial neural networks with
multiple layers to learn complex patterns from large datasets. It is particularly effective for
image recognition and analysis tasks like brain tumor detection.

 MRI (Magnetic Resonance Imaging): A medical imaging technique that uses strong
magnetic fields and radio waves to produce detailed images of organs and tissues within the
body, including the brain. MRI scans are commonly used for brain tumor detection.

 Autoencoder (AE): A type of neural network architecture used for unsupervised learning.
Autoencoders learn to compress input data into a lower-dimensional representation while
preserving essential features. In this project, autoencoders may be used for noise reduction
in MRI scans before tumor detection.

 Vision Transformer (ViT): A deep learning model architecture specifically designed for
image classification tasks. ViTs excel at capturing subtle variations in image data, making
them well-suited for brain tumor detection in MRI scans.

 Brain Tumor Detection: The process of identifying the presence and location of tumors in
the brain using medical imaging techniques like MRI scans. Early and accurate brain tumor
detection is crucial for effective patient care and treatment planning.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 2

1.4 Document Conventions

This section describes the standardized format and style guidelines followed throughout this
document. The document starts with a Title Page that includes the document's title, project name,
version number, date, and author(s). Following the Title Page, a Table of Contents is provided to list
all sections and subsections along with their corresponding page numbers for easy navigation.

To ensure clarity and consistency in specifying requirements, a uniform format is utilized. Each
requirement is identified by a unique identifier and includes a description, priority, source,
dependencies, and acceptance criteria. This structured approach helps in maintaining the
document’s readability and facilitates understanding by all stakeholders involved.

1.5 References and Acknowledgments

[1] Mukherjee, S., Dasgupta, S., & Sarkar, S. (2020). Evaluation of Pathway to Diagnosis of
Pediatric Brain Tumors in Tamil Nadu, India. Journal of Clinical Oncology, 38(18), 2082-
2082. doi: 10.1200/GO.23.00214

[2] Lamburn, Emily, et al. "MRI Reporting Times in a District General Hospital: A Retrospective
Audit." The British Journal of Radiology, vol. 87, no. 1042, 2014, pp. 20130447.
https://www.sciencedirect.com/science/article/abs/pii/S154614401931484X

[3] Dos Santos, Cicero F., et al. "How Powerful Are Vision Transformers?" arXiv preprint
arXiv:2010.11929 (2020).

[4] Yao, R., Wang, N., Chen, P. et al. A CNN-transformer hybrid approach for an intrusion
detection system in advanced metering infrastructure. Multimed Tools Appl 82, 19463–
19486 (2023).

[5] Babu Vimala, B., Srinivasan, S., Mathivanan, S.K. et al. Detection and classification of brain
tumor using hybrid deep learning models. Sci Rep 13, 23029 (2023).

[6] Verma, A., Gupta, N., Bhatele, P. et al. JMCD Dataset for Brain Tumor Detection and
Analysis Using Explainable Deep Learning. SN COMPUT. SCI. 4, 840 (2023).

[7] Younis, A.; Qiang, L.; Nyatega, C.O.; Adamu, M.J.; Kawuwa, H.B. Brain Tumor Analysis
Using Deep Learning and VGG-16 Ensembling Learning Approaches. Appl. Sci. 2022, 12,
7282. https:// doi.org/10.3390/app12147282
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 3

2 Overall Description
2.1 Product Perspective
The "Brain Tumor Detection using Vision Transformer" project aims to revolutionize medical imaging
analysis with the following key features:

 Advanced AI Technology: Utilizing Vision Transformer (ViT) for precise brain tumor
detection from medical images.
 Automated Diagnosis: Streamlining diagnostic workflows with automated detection
capabilities.
 User-Friendly Interface: Intuitive design for seamless integration into existing healthcare
systems.
 Scalability and Flexibility: Adaptable to various datasets and compatible with different
medical imaging modalities.

This perspective highlights the project's focus on enhancing diagnostic accuracy and efficiency in
clinical settings through innovative AI technology.

2.2 Product Functionality

System Architecture

Fig.1. System Architecture for Brain Tumor Detection

Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 4

The "Brain Tumor Detection using Vision Transformer" project centers on evaluating the robustness
of Vision Transformer (ViT) technology in classifying brain tumors from noisy medical imaging data.
The project involves collecting diverse MRI or CT scan images with varying degrees of noise and
preprocessing them to standardize inputs for ViT analysis. ViT's performance is rigorously tested in
real-world scenarios to accurately classify scans into categories such as notumor, meningioma,
glioma, and pituitary tumor. The study focuses on assessing ViT's capability to handle noise and its
effectiveness in differentiating between tumor types without evaluating size or other detailed
characteristics. This research aims to contribute insights into ViT's applicability in medical
diagnostics with noisy datasets, without direct integration into clinical settings.

2.2 Users and Characteristics

 Machine Learning Engineers:
 Need technical details on model training and integration.
 Require clear APIs, data flow diagrams, and system architecture details.

 Software Developers:
 Require clear APIs and integration documentation.
 Need information on data preprocessing, normalization, and model deployment.

 Data Scientists:
 Need information on data acquisition, cleaning, and model evaluation.
 Require access to datasets and insights into model performance metrics.
 Project Managers:
 Require high-level overviews, project timelines, and milestones.
 Need tools for tracking project progress and milestones.

 Quality Assurance Specialists:

 Need detailed test plans, test cases, and validation procedures.
 Require performance metrics and quality benchmarks.

 Users in Medical Field:

 Require a user-friendly interface.
 Clear instructions for system operation

2.3 Operating Environment

The "Brain Tumor Detection using Vision Transformer" project operates in a Jupyter Notebook
environment utilizing Python version 3.12. Key libraries include NumPy for efficient numerical
computations, pandas for data manipulation, matplotlib and seaborn for data visualization, and scikit-
learn for machine learning model implementations and evaluation. Additionally, Tkinter will be used
for developing a user-friendly desktop application, providing healthcare professionals with intuitive
access to brain tumor detection results. This integrated environment supports comprehensive
development and deployment of machine learning algorithms and desktop applications essential for
advancing diagnostic capabilities in medical imaging.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 5

2.4 Design and Implementation Constraints

Several critical design and implementation constraints are essential for developing the Brain Tumor
Detection using Vision Transformer model:

 Processing Efficiency: The system must efficiently analyze MRI or CT scan images using
Vision Transformer to ensure timely and accurate detection of brain tumors without delays in
clinical settings.

 Robustness to Noise: The model should be resilient to noise commonly found in medical
imaging data, such as artifacts or variations in image quality, to maintain high accuracy in
tumor detection across different imaging conditions.

 Scalability: The architecture should be scalable to handle diverse datasets and increasing
volumes of medical images, ensuring consistent performance and accuracy as the dataset
size grows.

These constraints are pivotal in designing a reliable and scalable system that can effectively detect
brain tumors in noisy medical imaging data, supporting accurate clinical diagnoses and treatments.

2.5 User Documentation

User documentation for the "Brain Tumor Detection using Vision Transformer" project aims to
provide comprehensive guidance for users interacting with the system. Key elements include:

 System Overview: A high-level introduction to the project, outlining its objectives and the
role of Vision Transformer in brain tumor detection.
 Installation Guide:
 Step-by-step instructions for setting up the required Python environment (Python
3.12) and installing necessary libraries (NumPy, pandas, matplotlib, seaborn, scikit-
learn, Tkinter).
 Usage Instructions:
 Detailed guidelines on running the Jupyter Notebook for image preprocessing, model
training, and inference.
 How to interact with the Tkinter-based desktop application for accessing and
visualizing brain tumor detection results.
 Troubleshooting:
 Common issues and troubleshooting steps for handling potential errors during setup
or operation.
 Additional Resources:
 Links to relevant research papers, documentation on Vision Transformer, and
datasets used in the project for further exploration.

This user documentation aims to empower users, including healthcare professionals and
researchers, with the knowledge and tools necessary to effectively utilize and understand the "Brain
Tumor Detection using Vision Transformer" system.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 6

2.6 Assumptions and Dependencies

Several assumptions and dependencies underpin the development and deployment of the "Brain
Tumor Detection using Vision Transformer" project:

 Data Quality: Assumes availability of sufficiently labeled and diverse MRI or CT scan
datasets for training and evaluation purposes.
 Hardware Requirements: Dependency on adequate computing resources capable of
handling intensive model training and inference tasks using Python 3.12, NumPy, pandas,
matplotlib, seaborn, scikit-learn, and Tkinter.
 Model Performance: Dependency on the robustness and accuracy of Vision Transformer
models in classifying brain tumor types under varying levels of noise and image quality.
 Integration with Tkinter: Assumes successful integration and functionality of the Tkinter-
based desktop application for user interface and result visualization.
 Maintenance and Support: Dependency on ongoing maintenance and support to address
potential updates in libraries, frameworks, or operating system changes affecting system
performance and functionality.

These assumptions and dependencies guide the project's planning and execution, ensuring
alignment with technical requirements and expected outcomes in advancing brain tumor detection
capabilities using Vision Transformer technology.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 7

3 Specific Requirements
3.1 External Interface Requirements
3.1.1 User Interfaces

The "Brain Tumor Detection using Vision Transformer" project features the following user interfaces:

 Jupyter Notebooks:
 Utilized for interactive data preprocessing, model training, and evaluation.
 Provides a flexible environment for data scientists and machine learning engineers to
experiment with different algorithms and configurations.

 Tkinter-based Desktop Application:

 Developed for healthcare professionals to interact with the system.
 Offers a user-friendly interface for uploading medical images, initiating tumor
detection processes, and visualizing classification results.
 Designed to facilitate intuitive navigation and provide clear, actionable insights into
brain tumor detection outcomes.

These interfaces aim to enhance user experience by offering both technical flexibility and intuitive
interaction capabilities tailored to different user roles within the project's scope.

3.1.2 Hardware Interfaces

The "Brain Tumor Detection using Vision Transformer" project requires a standard desktop or
laptop PC with the following specifications:

 Processor: Intel Core i5 or AMD Ryzen 5 (or equivalent) processor or higher.

 Memory: 8 GB RAM or higher.
 Storage: SSD storage recommended for faster data access.
 Graphics: Integrated graphics card or dedicated GPU recommended for faster image
processing in Jupyter Notebooks.
These specifications ensure the smooth operation of Jupyter Notebooks for data preprocessing,
model training, and evaluation tasks, as well as running the Tkinter-based desktop application for
user interface and visualization in the context of brain tumor detection.

3.1.3 Software Interfaces

The "Brain Tumor Detection using Vision Transformer" project interfaces with the following software
components:
 Python 3.12: Programming language used for implementing machine learning algorithms,
data preprocessing, and application development.
 Jupyter Notebooks: Interactive environment for data exploration, preprocessing, model
training, and evaluation using libraries such as NumPy, pandas, matplotlib, seaborn, and
scikit-learn.
 Tkinter: Python library for developing desktop applications with a graphical user interface
(GUI), utilized for visualizing brain tumor detection results and user interaction.
 Operating System Compatibility: Compatible with Windows, macOS, and Linux operating
systems to ensure broad accessibility and usability across different environments.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 8
3.1.4 Communications Interfaces

Since all components of the "Brain Tumor Detection using Vision Transformer" project are executed
on a local machine, there are no external communication interfaces required. The system operates
entirely within the local environment, encompassing data processing, model training, application
development using Python, and user interaction through Jupyter Notebooks and Tkinter-based
desktop application. This setup ensures efficient and secure data handling and application
performance without the need for external communication protocols or interfaces.

3.2 Functional Requirements

The "Brain Tumor Detection using Vision Transformer" project encompasses the following
functional requirements:

3.2.1 Image Acquisition

 Requirement: The system shall acquire high-quality MRI or CT scan images suitable for
brain tumor detection.
 Detail: Images will be sourced from medical imaging repositories or direct scans conducted
in clinical settings to ensure clarity and diagnostic relevance.

3.2.2 Image Preprocessing

 Requirement: The system shall preprocess images to optimize input for the Vision
Transformer model.
 Detail: This involves standardizing image sizes, enhancing contrast, and applying noise
reduction techniques to improve the clarity and interpretability of medical imaging data.

3.2.3 Tumor Classification

 Requirement: The system shall classify brain tumors into predefined categories (notumor,
meningioma, glioma, pituitary).
 Detail: Utilizing Vision Transformer technology, the system will analyze preprocessed
images to accurately categorize tumors based on learned patterns and features, providing
diagnostic insights.

3.2.4 Real-Time Processing

 Requirement: The system shall process each image efficiently, with processing times
optimized to support clinical decision-making.
 Detail: Processing speed will be calibrated to handle medical imaging data in real-time,
ensuring timely and responsive tumor detection and classification.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 9
3.2.5 Output Display

 Requirement: The system shall display brain tumor detection results through a Tkinter-
based graphical user interface (GUI).
 Detail: Results from tumor detection and classification will be visualized and interactively
presented on the Tkinter GUI, providing healthcare professionals with intuitive access to
diagnostic outcomes directly from the system.

These functional requirements highlight the core functionalities and user interactions of the "Brain
Tumor Detection using Vision Transformer" project, emphasizing its role in supporting efficient
and accurate brain tumor diagnosis using advanced machine learning techniques.

3.3 Behaviour Requirements

The "Brain Tumor Detection using Vision Transformer" project is governed by the following
behavior requirements:

 Real-time Processing: Process MRI or CT scan images and classify brain tumors within a
maximum of 2 seconds per image to support timely clinical decision-making.
 High Accuracy: Achieve a minimum brain tumor classification accuracy of 95% to ensure
reliable diagnostic outcomes.
 Reliability: Operate continuously without downtime, handling any system failures
gracefully to minimize disruption in medical imaging analysis.
 User-Friendly Interface: Provide an intuitive Tkinter-based graphical user interface (GUI)
for healthcare professionals, facilitating easy navigation and interpretation of brain tumor
detection results.
 Security: Ensure data privacy and integrity through encryption protocols and access
controls, safeguarding patient information and medical imaging data.
 Scalability: Support scalable deployment to accommodate varying volumes of medical
imaging data and operational demands, ensuring consistent performance and accessibility
across healthcare settings.

These behavior requirements outline the essential capabilities and operational standards for the
"Brain Tumor Detection using Vision Transformer" project, aiming to enhance diagnostic accuracy,
reliability, and user experience in clinical applications.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 10

4 Other Non-functional Requirements

4.1 Performance Requirements

The performance requirements for the "Brain Tumor Detection using Vision Transformer" project
are as follows:

1. Image Processing Speed:

The system should process each MRI or CT scan image for brain tumor detection within 2 seconds
to support real-time clinical decision-making.

2. Detection Accuracy:

Achieve a minimum classification accuracy of 95% in identifying different types of brain tumors
(notumor, meningioma, glioma, pituitary), validated against ground truth data and clinical
standards.

3. Scalability:

Initially support processing a minimum of 100 MRI or CT scan images per hour, with scalability to
handle increased throughput as medical imaging volumes grow.

4. Resource Utilization:

Optimize GPU and CPU utilization to efficiently handle computational tasks during peak demand
periods, ensuring stable performance without resource exhaustion.

5. Response Time:

Provide responsive interactions on the Tkinter-based GUI, maintaining a latency of under 500
milliseconds for displaying brain tumor detection results and facilitating user interactions.

6. Reliability:

Ensure system reliability with uptime of at least 99.9%, minimizing downtime and disruptions in
accessing and analyzing medical imaging data.

7. Training and Retraining:

Facilitate periodic model retraining within 24 hours of receiving new medical imaging data,
enhancing the system's ability to adapt and improve brain tumor detection accuracy over time.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 11

4.2 Safety and Security Requirements

The "Brain Tumor Detection using Vision Transformer" project prioritizes safety and security with
the following requirements:

1. Data Privacy:
 Ensure strict adherence to data privacy regulations (e.g., HIPAA) by implementing
encryption protocols for storing and transmitting medical imaging data.
2. Access Control:
 Implement robust access control mechanisms to restrict system access based on roles
and permissions, ensuring only authorized personnel can interact with sensitive
patient information.
3. System Integrity:
 Employ measures to maintain the integrity of medical imaging data and analysis
results, preventing unauthorized modifications or tampering.
4. Error Handling:
 Implement comprehensive error handling mechanisms to detect and mitigate errors
during data processing and analysis, minimizing risks to patient safety and data
integrity.
5. Continuity of Operations:
 Ensure continuity of system operations with backup and recovery procedures in
place to mitigate the impact of potential disruptions or data loss.
6. Ethical Considerations:
 Adhere to ethical guidelines in the use of AI and medical data, ensuring transparency
in how patient data is used for research and diagnostic purposes.
7. Auditability:
 Maintain audit trails of system activities and user interactions to facilitate
accountability and traceability of data access and modifications.

These safety and security requirements underscore the project's commitment to protecting patient
confidentiality, maintaining data integrity, and ensuring reliable and ethical use of AI technology in
medical diagnostics.
Software Requirements Specification for Brain Tumor Classification using Vision Transformer Page 12

4.3 Software Quality Attributes

The "Brain Tumor Detection using Vision Transformer" project adheres to the following software
quality attributes:

1. Accuracy:
o Ensure high accuracy in brain tumor detection with a minimum classification
accuracy of 95%, validated against ground truth data and clinical standards.
2. Reliability:
o Maintain system reliability with uptime of at least 99.9%, minimizing disruptions in
medical imaging analysis and clinical decision-making processes.
3. Performance Efficiency:
o Optimize image processing speed to analyze MRI or CT scan images within 2
seconds per image, supporting real-time diagnostic capabilities.
4. Usability:
o Provide a user-friendly Tkinter-based interface for healthcare professionals, ensuring
intuitive navigation and clear presentation of brain tumor detection results.
5. Security:
o Implement robust data encryption and access control measures to safeguard patient
confidentiality and prevent unauthorized access or data breaches.
6. Scalability:
o Support scalable deployment to accommodate increasing volumes of medical
imaging data and operational demands, ensuring consistent performance as the
system expands.
7. Maintainability:
o Facilitate ease of maintenance and updates by documenting code comprehensively,
enabling efficient troubleshooting, and supporting timely model retraining and
system enhancements.
8. Ethical Compliance:
o Adhere to ethical guidelines in AI and healthcare by ensuring transparency in data
usage, protecting patient privacy, and maintaining ethical standards in research and
clinical applications.

These software quality attributes collectively contribute to the reliability, usability, security, and
performance of the "Brain Tumor Detection using Vision Transformer" system, enhancing its
effectiveness in clinical settings for accurate and timely brain tumor diagnosis.