BRAIN CANCER DETECTION AND

CLASSIFICATION
SYNOPSIS

A Major Project submitted for the partial fulfillment of the degree of

Bachelor of Technology

in

Computer Science & Engineering

(Session 2023-2024)

Under the supervision

of Project Guide

Amrita Ticku

(Associate Professor)

Submitted by:

Saday Samnotra Avishkaar Chaturvedi Tanmay Sehgal Chayan Manocha

(75211502720) (35911502720) (35611502720) (35411502720)

Department of Computer Science & Engineering

BHARATI VIDYAPEETH’S COLLEGE OF

ENGINEERING PASCHIM VIHAR, NEW DELHI

1
February-2024

2
 “Detecting The Areas Affected In Human Body Due To The
Presence Of Brain Cancer Using CNN”

1. INTRODUCTION

In contemporary medical diagnostics, the advent of artificial intelligence (AI) technologies has
revolutionized the landscape of disease detection and identification[1]. Particularly, the realm of
neuro-oncology has witnessed profound advancements with the integration of AI algorithms into
magnetic resonance imaging (MRI) analyses for the detection and identification of brain cancer.
This scholarly discourse embarks upon a comprehensive exploration of the methodologies,
challenges, and prospects underlying the utilization of AI models, specifically the EfficientNet and
Vision Transformer models, in the detection and identification of brain cancer through MRI
scans[2].

a. Introduction to Brain Cancer and MRI Imaging

Brain cancer, or malignant brain tumors, constitutes a significant health challenge globally,
characterized by abnormal growths of cells within the brain tissue. The diagnosis and subsequent
treatment of brain cancer hinge critically upon accurate and timely detection of tumor presence,
size, and location. Magnetic resonance imaging (MRI) emerges as a cornerstone modality in
neuroimaging, offering unparalleled insights into the structural and functional attributes of the
brain. MRI employs powerful magnetic fields and radio waves to generate detailed images of the
brain's anatomy[3][4], facilitating the detection and characterization of abnormalities, including
tumors, with remarkable precision.

b. Integration of AI in Brain Cancer Detection

The integration of AI algorithms into MRI-based brain cancer detection endeavors represents a
paradigm shift in contemporary diagnostic methodologies[3][5][9]. AI models, endowed with
the capacity for pattern recognition and data interpretation, exhibit remarkable potential in
augmenting the accuracy and efficiency of tumor detection and identification. Among the
myriad AI architectures, the EfficientNet and Vision Transformer models stand out as
formidable contenders, each leveraging distinct computational paradigms to unravel the
intricacies of brain cancer pathology from MRI scans.

c. EfficientNet: A Framework of Scalability and Efficacy

EfficientNet, a seminal convolutional neural network (CNN) architecture, embodies the
principles of scalability and efficiency in neural network design. Proposed by Tan and Le
(2019), EfficientNet introduces a novel compound scaling method that systematically balances
model depth, width, and resolution to optimize performance across varying computational
constraints. Leveraging this framework, EfficientNet streamlines the process of feature
extraction from MRI images, enabling robust tumor detection and classification with minimal
computational overhead[1]

d. Vision Transformer: Unleashing the Power of Attention Mechanisms:

Contrasting the conventional CNN architectures, the Vision Transformer (ViT) model
propounded by Dosovitskiy et al. (2020) revolutionizes image processing through the utilization
3
of self-attention mechanisms. Inspired by the success of transformers in natural language
processing, ViT reimagines image analysis as a sequence-to-sequence translation task, wherein
image patches serve as input tokens subjected to self-attention operations.[10] By capturing
long-range dependencies and contextual relationships within MRI scans, Vision Transformer
transcends traditional CNN limitations, exhibiting prowess in discerning subtle tumor features
and nuances.

d. Ensemble Learning: Fusion of Model Expertise

Recognizing the complementary strengths of EfficientNet and Vision Transformer models,
ensemble learning emerges as a compelling strategy for harnessing the collective intelligence of
diverse AI architectures.[11] Through ensemble methods, such as averaging predictions or
employing stacking techniques, the individual predictions of multiple models are amalgamated
to yield a consolidated output with enhanced robustness and generalization capabilities. In the
context of brain cancer detection, ensemble learning mitigates the risk of model bias and
variance, fostering a holistic approach to diagnostic inference grounded in the consensus of
diverse AI perspectives.

e. Prospects and Challenges in AI-Enabled Brain Cancer Detection

While the integration of AI models holds immense promise in revolutionizing brain cancer
diagnostics, several challenges necessitate careful consideration. Foremost among these
challenges is the imperative for large-scale annotated datasets encompassing diverse tumor
phenotypes and imaging modalities to facilitate robust model training and validation. Moreover,
the interpretability of AI-driven diagnostic outputs remains a pressing concern, necessitating the
development of explainable AI frameworks to elucidate the rationale underlying model
predictions and instill confidence among clinicians and stakeholders.[11]

In conclusion, the fusion of AI models, namely EfficientNet and Vision Transformer, heralds a
new era of precision medicine in the realm of brain cancer detection and identification through
MRI scans[6]. By harnessing the collective intelligence of diverse computational architectures
and leveraging ensemble learning strategies, AI-enabled diagnostics offer unparalleled insights
into the nuances of tumor pathology, empowering clinicians with the tools necessary for early
intervention and personalized therapeutic strategies[7][8]. As research endeavors continue to
unravel the complexities of neuro-oncology, the synergy between AI and MRI imaging holds the
promise of ushering in transformative advancements in patient care and clinical outcomes.

4
 2. OBJECTIVES

1. Accessibility and inclusivity: Develop a system that can accurately detect and classify various
types of brain tumors from MRI images, enabling healthcare professionals to diagnose and treat
patients effectively.
2. Applications and integrations: Explore applications in medical diagnosis, treatment planning,
and research, with potential integrations into healthcare systems for seamless adoption by medical
practitioners.
3. To implement the Ensemble Learning model: Implement an ensemble model consisting of
EfficientNet and Vision Transformer architectures for enhanced brain tumor detection and
classification, and conduct a comparative analysis with previous models to evaluate performance
improvements.
4. Facilitate two-way communication: Enable the conversion of medical reports or spoken
descriptions into visual representations of brain tumors, enhancing communication between
healthcare providers and patients.
5. Research more on low resource languages: Investigate less common types of brain tumors and
rare neurological conditions, leveraging available data to improve detection and classification
accuracy in under-researched areas.
6. Create a web extension: Develop a web-based interface to allow medical professionals to access
and interact with the brain tumor detection system, integrating it seamlessly into their workflow.
7. Ease of learning: Provide user-friendly documentation and resources to educate medical
professionals and researchers on the utilization and interpretation of the brain tumor detection
system, promoting widespread adoption and understanding.

5
 3. MINOR OUTCOMES(COMPARATIVE ANALYSIS)

MINOR PROJECT MAJOR PROJECT

MODEL CNN Ensemble Learning (CNN + Efficient Net +

Vision Transformer)

FEATURES Classification Classification+Identification of area with

greater accuracy

DATASET 22 types of tumors with 22 types of tumors with 4000 images as

1500 images as training dataset.(3000 online and 1000 real-world)
dataset (1000 online and
500 real-world)

6
 4. PROPOSED METHODOLOGY

Brain cancer detection and identification are critical tasks in the field of medical diagnostics,
requiring advanced AI models to assist healthcare professionals in accurate and timely diagnosis.
The following proposed methodology outlines a comprehensive approach leveraging state-of-the-
art deep learning architectures and ensemble techniques for robust brain cancer detection and
identification.

1. Data Preprocessing and Augmentation:

➔ Begin by collecting a diverse and comprehensive dataset consisting of brain imaging scans,
such as MRI or CT scans, from multiple sources to ensure variability.
➔ Preprocess the dataset by standardizing image sizes, enhancing contrast, and normalizing
pixel intensities to ensure uniformity and improve model generalization.
➔ Augment the dataset using techniques like random flipping, zooming, and rotation to
increase the diversity of training samples and enhance model robustness against variations
in imaging conditions.

2. Model Selection and Architecture:

➔ Utilize pre-trained deep learning models from TensorFlow Hub, such as EfficientNet V2
and Vision Transformer (ViT), which have demonstrated exceptional performance in image
classification tasks.
➔ Fine-tune the selected models on the preprocessed dataset to adapt them specifically for
brain cancer detection, leveraging transfer learning to exploit features learned from large-
scale image datasets.
➔ Design custom classification heads with additional fully connected layers to capture
intricate patterns specific to brain cancer characteristics, ensuring the model's ability to
discern subtle differences.

7
3. Training and Evaluation:
➔ Divide the preprocessed dataset into training, validation, and testing sets, ensuring a
balanced distribution of samples across different classes to mitigate biases.
➔ Train the selected models using rigorous training protocols, including early stopping and
learning rate scheduling, to prevent overfitting and ensure convergence towards optimal
performance.
➔ Evaluate the trained models using comprehensive metrics such as accuracy, precision,
recall, F1-score, and area under the ROC curve (AUC) to assess their performance across
different evaluation criteria.
➔ Validate the models on a separate validation set to evaluate their generalization capability
and fine-tune hyperparameters based on validation performance.

4. Ensemble Techniques:
➔ Employ ensemble techniques to combine predictions from multiple individual models, such
as averaging or weighted averaging, to leverage diverse model architectures and enhance
predictive accuracy.
➔ Explore geometric mean-based ensembles to mitigate the risk of outlier predictions and
improve overall ensemble performance, particularly effective when combining predictions
from models with varying degrees of confidence.

5. Interpretation and Visualization:

➔ Employ interpretability techniques such as Grad-CAM or SHAP to visualize model
predictions and highlight regions of interest within brain imaging scans, aiding clinicians in
understanding the model's decision-making process.
➔ Generate comprehensive reports summarizing model performance, including confusion
matrices, ROC curves, and precision-recall curves, to provide detailed insights into the
model's strengths and limitations.

6. Deployment and Integration:

➔ Deploy the trained ensemble model in a secure and scalable environment, integrating it into
existing healthcare infrastructure for seamless integration into clinical workflows.
➔ Implement continuous monitoring and quality assurance measures to ensure model
performance consistency and reliability over time, facilitating continuous improvement and
adaptation to evolving healthcare needs.

The proposed methodology combines cutting-edge deep learning architectures, rigorous training
protocols, and ensemble techniques to develop a robust AI model for brain cancer detection and
identification. By leveraging the collective strengths of individual models and employing
sophisticated evaluation metrics, interpretability techniques, and deployment strategies, the
proposed approach aims to deliver accurate, reliable, and clinically actionable insights to healthcare
professionals, ultimately enhancing patient outcomes and advancing the field of medical
diagnostics.

8
5.TIME FRAME REQUIRED FOR VARIOUS STAGES OF
IMPLEMENTATION (GANTT CHART)

9
 6. REFERENCES

[1] S. Banerjee, S. Mitra and B. U. Shankar, "Synergetic neuro-fuzzy feature selection and
classification of brain tumors," 2017 IEEE International Conference on Fuzzy Systems (FUZZ-
IEEE), Naples, Italy, 2017, pp. 1-6, doi: 10.1109/FUZZ-IEEE.2017.8015514. keywords: {Feature
extraction;Tumors;Pragmatics;Magnetic resonance imaging;Cancer;Shape},

[2] M. Gurbină, M. Lascu and D. Lascu, "Tumor Detection and Classification of MRI Brain
Image using Different Wavelet Transforms and Support Vector Machines," 2019 42nd International
Conference on Telecommunications and Signal Processing (TSP), Budapest, Hungary, 2019, pp.
505-508, doi: 10.1109/TSP.2019.8769040. keywords: {Tumors;Support vector machines;Magnetic
resonance imaging;Noise reduction;Discrete wavelet
transforms;brain;classification;denoising;detection;support vector machines;tumor;wavelet
transforms},

[3] M A Ansari, Rajeev Agrawal, and Rajat Mehrotra, “Detection and classification of brain
tumor in MRI images using wavelet transform and support vector machine,” 2002.

[4] A. Hebli and S. Gupta, "Brain tumor prediction and classification using support vector
machine," 2017 International Conference on Advances in Computing, Communication and Control
(ICAC3), Mumbai, India, 2017, pp. 1-6, doi: 10.1109/ICAC3.2017.8318767. keywords:
{Tumors;Feature extraction;Support vector machines;Discrete wavelet
transforms;Training;Magnetic resonance imaging;Principal component analysis;Brain Tumor;MRI
(magnetic resonance imaging);DWT (Discrete wavelet transform);PCA (Principal component
analysis);SVM (Support vector machine)},

[5] J. Amin, M. Sharif, A. Haldorai, M. Yasmin, and R. S. Nayak, "Brain tumor detection and
classification using machine learning: a comprehensive survey," Complex & Intelligent Systems,
vol. 8, pp. 3161-3183, 2021.

[6] T. Sadad, A. Rehman, A. Munir, Tanzila, U. Tariq, Noor, and R. A. Rehman, "Brain tumor
detection and multi-classification using advanced deep learning techniques,". Available:
[https://www.semanticscholar.org/paper/Brain-tumor-detection-and-classification-using-a-Amin-
Sharif/f630ee33272b8505668b70975c3c9df015b21105].

[7] Y. Bhanothu, A. Kamalakannan and G. Rajamanickam, "Detection and Classification of

Brain Tumor in MRI Images using Deep Convolutional Network," 2020 6th International
Conference on Advanced Computing and Communication Systems (ICACCS), Coimbatore, India,
2020, pp. 248-252, doi: 10.1109/ICACCS48705.2020.9074375. keywords: {Tumors;Classification
algorithms;Proposals;Image segmentation;Training;Magnetic resonance imaging;Testing;Brain
tumor detection;glioma;meningioma;pituitary;Faster R-CNN;VGG-16;mean average
precision;MRI},
10
[8] D. M. Joshi, N. K. Rana and V. M. Misra, "Classification of Brain Cancer using Artificial
Neural Network," 2010 2nd International Conference on Electronic Computer Technology, Kuala
Lumpur, Malaysia, 2010, pp. 112-116, doi: 10.1109/ICECTECH.2010.5479975. keywords:
{Artificial neural networks;Neoplasms;Magnetic resonance imaging;System testing;Cancer
detection;Feature extraction;Computer networks;Lesions;Image processing;Histograms;Brain
cancer;MRI;Co-occurrence Matrix;ANN},

[9] H. T. Zaw, N. Maneerat and K. Y. Win, "Brain tumor detection based on Naïve Bayes
Classification," 2019 5th International Conference on Engineering, Applied Sciences and
Technology (ICEAST), Luang Prabang, Laos, 2019, pp. 1-4, doi: 10.1109/ICEAST.2019.8802562.
keywords: {Tumors;Magnetic resonance imaging;Image segmentation;Feature
extraction;Cancer;Classification algorithms;Databases;Glioblastoma Multiforme (GBM);Magnetic
resonance imaging (MRI);Naïve Bayes classification;maximum entropy threshold;statistical
features extraction},

[10] P. Thirumurugan and P. Shanthakumar, "Brain tumor detection and diagnosis using ANFIS
classifier," International Journal of Imaging Systems and Technology, vol. 26, 2016. [Online].
Available: https://api.semanticscholar.org/CorpusID:40434958

[11] S. Chauhan, A. More, R. Uikey, P. Malviya and A. Moghe, "Brain

tumor detection and classification in MRI images using image and data
mining," 2017 International Conference on Recent Innovations in Signal
processing and Embedded Systems (RISE), Bhopal, India, 2017, pp. 223-
231, doi: 10.1109/RISE.2017.8378158. keywords: {Feature
extraction;Tumors;Image segmentation;Image color analysis;Image edge
detection;Data mining;Clustering algorithms;MRI image;L∗a∗b colour
space;K-means clustering;segmentation;Classification},

11