Department of Electronics and Communication Engineering

Course name /Course code: R19EC381 / Innovative multidisciplinary Project

Title of the project : Brain Tumor Detection using ML

04/26/2025
 Class / Section / Semester : III ECE / (A) / VI
Batch : 2022-2026
Project Review : Second Review

Date : 01.04.2025
Project Guide with Designation : Dr. Dhanasekar.S AP/ECE

Name of the Industry ( if applicable) : AI & ML

Team Members : Bharathi A (722822106016)

Charu leela L (722822106020)
Harini K (722822106041)
Harish M (722822106045)

04/26/2025
 CHALLENGE STATEMENT
• Brain tumors are a serious medical condition with high mortality rates if not detected
and treated early. Traditional diagnosis methods, such as MRI and CT scans, rely on
manual assessment by radiologists, which can be subjective, timeconsuming, and prone
to human error.
• Additionally, distinguishing between different tumor types and non-tumorous
abnormalities requires high expertise and precision. To develop an AI-powered system
that can accurately detect and classify brain tumors from medical imaging data.
• By integrating Artificial Intelligence (AI) and Machine Learning (ML), the system will
automate the analysis of MRI scans, identifying tumors, segmenting affected regions,
and classifying tumor types
• The goal is to assist radiologists by providing faster, more consistent, and more precise
diagnoses, improving patient outcomes and treatment planning.

04/26/2025
 CONCEPT / SCOPE OF SOLUTION
• Brain tumor classification is a critical task in medical imaging, requiring precise
feature extraction and spatial understanding for accurate diagnosis.
• In this work, we propose a novel CNN-ViT-GNN hybrid model that integrates
Convolutional Neural Networks (CNNs) for local feature extraction, Vision
Transformers (ViTs) for global contextual learning, and Graph Neural Networks
(GNNs) to model tumor spread within brain regions.
• Our model leverages multi-scale feature fusion to enhance classification accuracy
while preserving spatial relationships between tumor regions.
• Additionally, the integration of GNN enables better tumor spread analysis,
improving interpretability in clinical applications. The proposed architecture offers
a computationally efficient and robust solution for brain tumor classification,
paving the way for more precise and reliable neuro-oncological diagnostics.
04/26/2025
 LITERATURE SURVEY / BACKGROUND STUDY
S. TITLE AUTHOR NAME OF DESCRIPTION
No ORGANIZATION
&YEAR
1. Brain tumor detection Khaliki, M.Z., Brain tumor detection • In this study, we aim to classify brain tumors
from images and Başarslan. from images and such as glioma, meningioma, and pituitary
comparison with transfer comparison with transfer tumor from brain MR images. Convolutional
learning methods and 3- learning methods and 3- Neural Network (CNN) and CNN-based
layer CNN layer CNN. Sci Rep 14, inception-V3, EfficientNetB4, VGG19,
2664 (2024). transfer learning methods were used for
https://doi.org/10.1038/s classification.
41598-024-52823-9

2. An efficient brain tumor Rao, K. Nishanth et Heliyon, Volume 10, • The proposed approach employing ResNet50
detection and al. Issue 17, e36773 and EfficientNet demonstrated higher levels
classification using pre- of accuracy, precision, and recall in detecting
trained convolutional brain tumors.
neural network models

Minimum 6 papers

04/26/2025
 LITERATURE SURVEY / BACKGROUND STUDY
S. TITLE AUTHOR NAME OF DESCRIPTION
No ORGANIZATION
&YEAR
3. Classification of Evangelia I. Wiley Online • The objective of this study is to provide an
brain tumor type and Zacharaki, Sumei Library,Volume 62, automated tool that may assist in the imaging
grade using MRI Wang, Sanjeev Issue 6, evaluation of brain neoplasms by
texture and shape in a Chawla, Dong Soo Published online: determining the glioma grade and
machine learning Yoo, Ronald Wolf, 26 October 2009 differentiating between different tissue types,
scheme Elias R. Melhem, such as primary neoplasms (gliomas) from
Christos Davatzikos secondary neoplasms (metastases).

4. Image Analysis for MRI Nilesh Bhaskarrao https://doi.org/ • In this study, to improve the performance and
Based Brain Tumor Bahadure, Arun 10.1155/2017/9749108 reduce the complexity involves in the
Detection and Feature Kumar Ray, Har Pal medical image segmentation process, we
Extraction Using Thethi have investigated Berkeley wavelet
Biologically Inspired transformation (BWT) based brain tumor
BWT and SVM segmentation.
Minimum 6 papers

04/26/2025
 EXISTING METHOD
Traditional Methods:

•MRI & CT Scans: High-resolution imaging; requires manual assessment by radiologists, which
can be subjective.
•Biopsy: Gold standard for confirmation; invasive with risks of infection and complications.
•Histopathological Analysis: Microscopic tissue examination; time-consuming and dependent on
expert pathologists.

Machine Learning-Based Methods:

•CNNs: Extracts features from MRI images; needs large datasets and high computational power.
•Transfer Learning: Uses pretrained models for better efficiency; prone to overfitting on small
datasets.
•SVMs: Classifies tumors based on texture; requires manual feature engineering.
•Hybrid Models: Combines CNN, ViT, and GNN for better accuracy; computationally expensive.

04/26/2025
 PROPOSED / INNOVATIVE METHOD

•Hybrid CNN-ViT-GNN Model: Combines CNN for local feature extraction,

ViT for global spatial learning, and GNN for tumor spread modeling.

•Feature Fusion: Attention-based mechanism integrates extracted features for

enhanced classification accuracy.

•Improved Interpretability: GNN enhances understanding of tumor structure

and progression.

•Higher Accuracy & Efficiency: Reduces false positives, improves diagnosis

speed, and aids radiologists in precise decision-making

04/26/2025
 BLOCK / SCHEMATIC DIAGRAM

04/26/2025
 USE CASE DIAGRAM

04/26/2025
 SPECIFICATION & BUDGET
S.No HARDWARE COMPONENTS SPECIFIATIONS COST

Total xxxx

S.No SOFTWARE SPECIFICATIONS

1. Python - programming language

2. ML framework - scikit learn , numpy , pandas

3. Image processing - open cv

04/26/2025
 COMPLETED WORKS UPTO THIS REVIEW

• Video (using hyperlink) , photograph ….can be added

This CNN model using ResNet-50 algorithm traces the brain tumor with the percentage of area it
has affected (an existing model implemented to check the likelihood)

04/26/2025
• Left Image (Normalized MRI Slice): A grayscale MRI slice of the brain.
• Right Image (Tumor Mask): A segmentation mask where red regions indicate tumor
areas, with a distinct white core possibly representing the tumor’s most aggressive part.
 The image displays different MRI sequences (FLAIR, T1, T1CE, and T2), each highlighting
unique tumor characteristics. The segmentation mask visually differentiates tumor regions, aiding
in precise tumor identification and analysis.

04/26/2025
 OUTPUT VIDEO

04/26/2025
 EXPERIMENTAL RESULTS

04/26/2025
 CONCLUSION
• Brain tumor detection has evolved from traditional methods like MRI, CT scans, and
biopsies to advanced machine learning techniques. While conventional approaches
provide accuracy, they are time-consuming and subjective.

• Machine learning-based methods, especially CNNs, SVMs, and hybrid models,

enhance detection accuracy and efficiency. However, challenges like data scarcity, high
computational costs, and interpretability issues remain.

• The proposed method aims to address these limitations by integrating deep learning
and innovative techniques, improving early diagnosis and patient outcomes. Future
advancements will focus on refining model accuracy, reducing false positives, and
enhancing clinical adoption.

04/26/2025
 PENDING WORK
DESCRIPTION EXPECTED DUE DATE REASON FOR
PENDING
Combination of GCN+ViT to 13/04/2025 Due to GPU usage
the prepared hybrid model limit and module
installation issues

04/26/2025
 CONTEST PARTICIPATION

Participated in the above Hackathon.

04/26/2025
 REFERENCES

• https://www.cell.com/heliyon/fulltext/S2405-8440(24)12804-6
• https://www.nature.com/articles/s41598-024-52823-9
• https://onlinelibrary.wiley.com/doi/full/10.1002/mrm.22147
• https://onlinelibrary.wiley.com/doi/full/10.1155/2017/9749108

04/26/2025
 THANK YOU

04/26/2025