COM748 Masters Research Project

Project Proposal Form

Student Name Student Number Supervisor Name

Project Title
Brain Tumour detection using CNN machine learning

Problem Statement (100 words)

Brain tumours are a serious health challenge, disrupting essential brain functions like waste
removal and nutrient absorption, and leading to complications such as kidney disease and vascular
issues. These conditions can worsen quickly, often with limited warning, making early detection vital
for improving patient outcomes. However, reliable patient data is lacking, and current tools struggle
with consistent, early identification. This project proposes creating a deep learning-based tool to
detect and monitor brain tumours in real-time, allowing healthcare providers to catch and respond to
changes early. With this approach, we aim to offer a transformative, accessible solution for timely
diagnosis.

Background Research (500 words)

Brain tumours are abnormal cell growths within the brain or surrounding tissues. They are classified
into various types based on their origin, malignancy, and growth patterns. Gliomas, for instance, are
derived from glial cells, which support and protect neurons. These tumours constitute about 30% of
all brain and central nervous system tumours, with two main categories: low-grade gliomas (Grade I
or II) and high-grade gliomas (Grade III or IV). Low-grade gliomas typically grow slowly, while high-
grade gliomas, such as glioblastoma multiforme (GBM), are highly aggressive and often fatal due to
their rapid growth and ability to invade surrounding brain tissue type of brain tumour is the
astrocytoma, which originates from astrocytes, star-shaped cells that form part of the brain's
supportive tissue. Astrocytoma is further classified into low-grade and high-grade categories. The
latter includes anaplastic astrocytoma, which tend to have aggressive growth patterns and are
difficult to surgically remove due to their invasive nature. Meningiomas from the meninges—
protective membranes surrounding the brain—are typically benign, but their size and location can
lead to significant health problems, requiring surgical removal. Pituitary tumours, all and often benign,
can interfere with hormonal regulation and cause various health issues due to their location at the
brain's base.

Accurate diagnosis of brain tumours relies on imaging, particularly MRI scans, which can identify
structural abnormalities. However, distinguishing between malignant and benign tumours, as well as
identifying the type of tumour, presents significant challenges. This is where Artificial Intelligence (AI)
can make a crucial impact. AI, particularly machine learning algorithms, is revolutionizing the way
medical imaging is analysed. In recent years, AI has shown great promise in aiding the detection and
classification of brain tumours from MRI scans. The use of deep learning and other advanced
techniques has proven to be effective in distinguishing between different tumour types, enabling
earlier detection and more personalized treatment plans.

AI's integration into healthcare dates to the 1950s, facial intelligence" coined in 1956. Early AI
research focused on symbolic reasoning and problem-solving. However, significant strides were
made in the 1960s and 1970s when the U.S. Department of Défense funded projects to develop
computers capable of mimicking human cognitive functions. Since then, AI has been incorporated
into diverse fields, including healthcare, where it has been used for tasks ranging from administrative
duties to medical diagnosis. In medical imaging, AI has found its most impactful applications,
especially y. It assists radiologists by automatically identifying patterns and anomalies in medical

1
images, leading to faster and more accurate diagnoses.

In the context of brain tumour detection, AI algorithms, particularly those based on *deep, can
analyse MRI scans for subtle changes that may not be visible to the human eye. These tools have
demonstrated success in identifying gliomas, astrocytoma, and meningiomas, and can even
differentiate between benign and malignant tumours. The integration of AI in medical imaging holds
the potential to reduce diagnostic errors, speed up diagnosis, and ultimately improve patient
outcomes.

As the field progresses, research is focused on improving the accuracy and reliability of AI models
like data diversity and model interpretability. Furthermore, AI’s role in healthcare continues to
expand, with a particular emphasis on personalized medicine, where AI-driven insights enable more
tailored treatments for individual patients. This project, focused on developing AI-based systems for
brain tumour classification from MRI images, builds on these advancements to improve the accuracy
and efficiency of brain tumour diagnosis and treatment. words?

Project Aim and Objectives (300 words)

The aim of this research is to develop a deep learning system utilizing Convolutional Neural
Networks (CNNs) for the accurate detection and classification of brain tumours, specifically chronic
brain tumour diseases, from MRI images. The system will be designed to handle imbalanced
datasets and ensure reliable performance even with unseen data, ultimately assisting healthcare
professionals in diagnosing and classifying brain tumours more efficiently and accurately.

Project Objectives:

1. Data Collection and Preprocessing:

 Collect MRI datasets, primarily from Kaggle, containing brain tumour images.
 Preprocess the data to standardize image dimensions, address data imbalances using
techniques such as oversampling or augmentation and improve image quality for better
model performance.

2. Model Development:
 Design and implement a CNN architecture tailored for brain tumour classification,
leveraging modern deep learning frameworks and GPU-based processing to optimize
the image analysis process.
 Train the CNN model on the processed dataset to classify different brain tumour
types, including benign and malignant tumours.

3. Model Evaluation:
 Evaluate the model using classification metrics such as accuracy, precision, recall,
and F1-score, focusing on its performance with imbalanced data and its ability to
accurately differentiate tumour types.
 Optimize the model through techniques like hyperparameter tuning, regularization,
and dropout to enhance its generalization.

4. Testing with Unseen Data:

 Test the model on unseen MRI images to evaluate its robustness and ensure its
effectiveness in real-world scenarios.

5. Application Development:
 Develop a user-friendly application that integrates the trained CNN model for
automated brain tumour classification, supporting healthcare professionals in clinical
decision-making.

This project aims to advance AI-based brain tumour detection systems, enhancing early diagnosis
and improving patient outcomes.

2
Methodology/approach (400 words)
Methodology/Approach

Methodology including data collection, preprocessing, model design, training, and evaluation, using
established methods in the field of medical imaging and machine learning.

1. Data Collection and Preprocessing

The first stage involves gathering MRI scan datasets that contain labelled images of brain tumours.
These datasets, typically sourced from open-access platforms like Kaggle, contain various tumour
types such as gliomas, astrocytoma, and meningiomas. To handle fewer samples, data
augmentation techniques, such as rotating, flipping, and zooming images, are applied to artificially
increase the size and diversity of the dataset, ensuring better generalization of the model.

This involves converting images to grayscale, applying histogram equalization for better contrast,
and thresholding to highlight tumour features. Any noise in the images is also reduced using filtering
techniques, which help in improving the overall quality of the input data.

2. Model Design and Development

A Convolutional Neural Network (CNN) is used as the core model architecture due to its proven
ability in image recognition tasks. The CNN model is designed with multiple convolutional layers to
extract features from the images, followed by pooling layers to reduce dimensionality and fully
connected layers for classification. The model will employ an activation function, such as ReLU, to
introduce non-linearity and improve learning. To optimize the model’s performance,
hyperparameters such as the learning rate, batch size, and number of epochs will be fine-tuned.

3. Training and Validation

For the model’s training, a supervised learning approach is adopted, where the labelled data is used
to teach the model how to recognize and classify brain tumors. The model’s performance is
validated using cross-validation techniques, which ensure that the model generalizes well to unseen
data. Additionally, the dataset is split into training, validation, and test sets to evaluate the model's
accuracy and robustness.

4. Evaluation and Testing

Once the model is trained, its accuracy and performance are assessed using various metrics,
including precision, recall, F1-score, and accuracy. The model is tested on a separate test dataset,
which contains previously unseen brain scans to measure its ability to make accurate predictions. A
confusion matrix is also generated to visualize the performance and identify any misclassifications.

5. Software and Hardware

For the implementation of the CNN model, Python, along with deep learning libraries like
TensorFlow and Keras, is used. These frameworks provide the necessary tools to build, train, and
evaluate the neural network. The training process is computationally intensive, and therefore, the
system is run on hardware with sufficient GPU capabilities to accelerate model training.

Legal/ Social/ Ethical Issues (200 words)

In this research project, several legal, social, and ethical issues must be considered to ensure that

3
the work adheres to the professional standards and codes of conduct in Computer Science.

Legal Considerations:

This project makes use of medical image datasets, primarily MRI scans, which often contain
sensitive health information. It is crucial to ensure that the datasets used are publicly available,
anonymized, and comply with privacy laws such as the General Data Protection Regulation (GDPR)
in the European Union or Health Insurance Portability and Accountability Act (HIPAA) in the United
States. The research must avoid using identifiable patient data without proper consent, ensuring
that all data handling and storage processes are secure and lawful.

Social Considerations:

The research has the potential to positively impact healthcare systems by improving the speed and
accuracy of brain tumour detection. However, the deployment of such technology in clinical settings
must consider potential social implications, such as accessibility and bias in the model. If the
dataset is not diverse enough, there could be a risk of inaccurate predictions for underrepresented
populations. This must be addressed by using a balanced and comprehensive dataset.

Ethical Considerations:

Ethically, the project must prioritize the well-being of individuals who might be impacted by the
results. This includes ensuring that the AI system does not provide harmful or incorrect diagnoses.
Ethical approval will be sought for the use of medical data, and the project will follow the ACM Code
of Ethics, ensuring transparency, fairness, and accountability in the research process. All
participants involved, even indirectly, will be respected, and data privacy will be prioritized.

Project & Risk Management (300 words)

Managing risks is a crucial aspect of ensuring the success of the brain tumor detection project
using deep learning techniques. Three primary types of risks must be considered: health and
safety risks, management risks, and technical risks.
1) Health and Safety Risks:
This project primarily involves computational tasks and does not entail physical risks related
to electrical equipment or hazardous environments. The main health concern is the potential
for strain from long hours of sitting and using computers. To mitigate this, regular breaks will
be scheduled, and ergonomic workstations will be used to minimize discomfort and prevent
repetitive strain injuries.
2) Management Risks:
A significant management risk involves completing the project within the given timeframe
while ensuring it meets all the required standards. If the software is developed too late, there
might not be enough time for proper testing and evaluation. To mitigate this risk, a detailed
project schedule will be created, including specific milestones for dataset preparation, model
design, and evaluation. Regular meetings with the supervisor will be held to monitor progress
and address any issues early. Additionally, contingency plans will be developed for potential
delays.
3) Technical Risks:
The technical risks for this project revolve around data quality, the effectiveness of the
convolutional neural network (CNN), and hardware limitations. If the dataset is incomplete or of
poor quality, it could hinder the model’s performance. To mitigate this, data cleaning and
augmentation techniques will be applied, and validation with external datasets will ensure
robustness. Another risk is the reliance on GPU availability for training deep learning models. If
hardware resources are limited, training time could be significantly impacted. To address this,
cloud-based computing platforms, such as Google Collab or AWS, will be used to ensure
adequate processing power.

4
References in IEEE format (Approx 5 References)

1. A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet classification with deep convolutional

neural networks," in *Proc. of the 25th Int. Conf. on Neural Information Processing Systems
(NIPS)*, Lake Tahoe, NV, USA, Dec. 2012, pp. 1097-1105.

2. D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," in *Proc. of the 3rd Int.
Conf. on Learning Representations (ICLR)*, San Diego, CA, USA, May 2015. [Online]. Available:
https://arxiv.org/abs/1412.6980.

3. L. S. B. Wu, "Brain tumor detection and classification using convolutional neural networks,"
*Journal of Medical Imaging and Health Informatics*, vol. 9, no. 3, pp. 511-517, Mar. 2019. doi:
10.1166/jmihi.2019.2653.

4. Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," *Nature*, vol. 521, no. 7553, pp. 436-444,
May 2015. doi: 10.1038/nature14539.

5. M. D. Zeiler and R. Fergus, "Visualizing and understanding convolutional networks," in *Proc. of

the European Conf. on Computer Vision (ECCV)*, Zurich, Switzerland, Sep. 2014, pp. 818-833.

These references cover fundamental concepts of CNN, optimization algorithms, and their
application in medical image processing, fitting well with the techniques used in this brain tumor
detection project.