SYNOPSIS FOR FINAL YEAR MAJOR PROJECT

“MRI Based Brain Tumor

Image Segmentation Model”

Bachelor of Engineering
In
Computer Science and Engineering

Proposed By

Samarth Garg & Kajal Singh

Enrolment No. A50105221018/ A50105221146

Under the guidance of

Dr. Ekta Soni

Department of Computer Science & Engineering

Amity School of Engineering & Technology
Amity University, Haryana
 AIM AND OBJECTIVE OF THE PROJECT

Aim:
The primary aim of this Computer Science and Engineering (CSE) project is to design, implement,
and evaluate a machine learning model. The project aims to leverage the power of data-driven
algorithms and deep learning models to create an accurate and efficient MRI Based Brain Tumor
Image Segmentation Model.

Objectives:

1. Data Collection and Preprocessing:

- Gather a comprehensive dataset containing relevant features related to the project.
- Clean and preprocess the dataset to ensure data quality.

2. Image Processing & Segmentation:

- Image Processing involves the manipulation and analysis of images using digital computers.
- Image Segmentation is a fundamental task in image processing and computer vision.

3. Deep Learning:
- Part of machine learning and AI that mimics the way humans acquire specific type of knowledge.
- Select and engineer relevant features that contribute to the predictive accuracy of the model.
- Normalize or scale features as necessary to ensure a consistent input for the machine learning
model.

4. Model Selection and Implementation:

- Choose machine learning model for image segmentation.
- Implement the model using appropriate libraries or frameworks, ensuring scalability and
efficiency.

5. Neural Networks:
- Computational Model inspired by neural structures and functions of human brain.
- Train the deep learning model on the training set and optimize hyperparameters for improved
performance.

6. Modules & Libraries:

 - Define appropriate evaluation metrics to assess the accuracy and reliability of the model
predictions.

7. Documentation and Reporting:

- Document the entire project, including data sources, methodology, implementation details, and
results.
- Prepare a comprehensive report summarizing the project, its findings, and the performance of the
developed model.
 BACKGROUND STUDY

Brain tumors, whether benign or malignant, require early detection and precise segmentation for
proper diagnosis and treatment. MRI (Magnetic Resonance Imaging) is the go-to method for this,
thanks to its detailed images of soft tissues. With different sequences like T1-weighted, T2-weighted,
and FLAIR, MRI scans help highlight and map out tumor areas in the brain. However, manually
segmenting these tumors is a tough, time-consuming job, often relying on radiologists to carefully
trace the tumor boundaries, which can be subjective and prone to mistakes.

Deep learning, especially through Convolutional Neural Networks (CNNs), has become a game
changer in automating this process. These models can learn complex patterns from large MRI
datasets, enabling them to detect tumors with much higher precision than traditional methods. But
building an effective deep learning model for brain tumor segmentation isn’t without its challenges.
Tumors vary greatly in size, shape, and texture, and the imaging quality can differ from patient to
patient, making it a complex task. On top of that, the limited availability of annotated medical datasets
makes training these models even more difficult.

Despite these hurdles, deep learning has made huge strides in medical imaging. Advanced models
like U-Net, 3D CNNs, and attention mechanisms have boosted the performance of segmentation
models, making it easier to automatically detect tumors, classify them, and assist doctors in making
better decisions. As this technology continues to improve, MRI-based brain tumor segmentation using
deep learning could truly revolutionize how we approach medical diagnostics.
 METHODOLOGY

The methodology for the project involves a systematic approach to developing a machine learning
model using deep learning.

1. Data Collection and Preprocessing:

- Acquire a comprehensive dataset containing relevant features.
- Conduct exploratory data analysis (EDA) to identify correlations between features.
- Perform data preprocessing, encoding categorical variables, and scaling numerical features.

2. Image Processing & Segmentation:

- Image Processing involves the manipulation and analysis of images using digital computers.
- Image Segmentation is a fundamental task in image processing and computer vision.

3. Deep Learning:
- Part of machine learning and AI that mimics the way humans acquire specific type of knowledge.
- Select and engineer relevant features that contribute to the predictive accuracy of the model.
- Normalize or scale features as necessary to ensure a consistent input for the machine learning
model.

4. Model Selection and Implementation:

- Choose machine learning model for image segmentation.
- Implement the model using appropriate libraries or frameworks, ensuring scalability and
efficiency.

5. Neural Networks:
- Computational Model inspired by neural structures and functions of human brain.
- Train the deep learning model on the training set and optimize hyperparameters for improved
performance.

6. Modules & Libraries:

- Define appropriate evaluation metrics to assess the accuracy and reliability of the model
predictions.

7. Documentation and Reporting:

 - Document the entire process, including data preprocessing steps, feature engineering, model
selection, training details, and evaluation results.
- Prepare a comprehensive report summarizing the methodology, key findings, and insights gained
from the deep learning model.

Project Stages (Flowchart)

1.Importing Libraries and Dataset:
o In this section we will load a few libraries which we will need to develop, visualize and test
our model. We will also be loading our dataset for one of the imported libraries named
Sklearn.
o Import the stated libraries: - NumPy - Pandas - Sklearn - matplotlib – Seaborn

2.Exploring and Preprocessing the Dataset:

In this section we will analyse our dataset using different methods and then we'll create a data frame
using the same. We will also carry out preprocessing on the data frame.

3.Model Implementation:
In this section we will import deep learning model from Sklearn. Use features identified from heatmap
and label to create training and testing set. Finally, we will train our model using training set.

4.Model Testing:
In this section we will test our prediction with testing data and calculate score to measure model
accuracy. We will also plot the results of the machine learning model.

Importing Exploring and Model

Model
Libraries and Preprocessing Testing
Implementation
Datasets the Dataset
TOOLS AND TECHNIQUES TO BE USED

The main technologies used are:

• Python
• Google Collab
• Libraries and Frameworks
o NumPy and Pandas: For data manipulation and preprocessing.
o Scikit-learn: For implementing the linear regression model and other machine learning
utilities.
o Matplotlib and Seaborn: For data visualization.
o Statsmodels: Useful for detailed statistical analysis of the model.

Hardware configuration:
Processor: 1.3 GHz.
RAM: 128 MB
Free Space required on hard disk: 10 MB

Software requirement:
Google collab
NumPy, Pandas, Matplotlib and Seaborn, Scikit-learn
 PROPOSED WORK

In this project, we aim to build a deep learning model that can automatically segment brain tumors
from MRI scans, making the process faster and more accurate. We'll start by using publicly available
datasets like the BraTS dataset, which includes various MRI sequences (such as T1, T2, and FLAIR)
along with manually labeled tumor regions. To prepare the data for the model, we’ll clean and process
the images normalizing them, removing irrelevant parts of the scans (like the skull), and resizing them
for uniformity. We'll also augment the data by flipping, rotating, and zooming the images to increase
variety and improve model performance.

The core of our model will be a Convolutional Neural Network (CNN) built specifically for
segmentation tasks. We’ll use U-Net architecture, which is great for this kind of work because it can
capture both the broad context and the fine details needed to accurately outline tumors. Since MRI
scans consist of multiple slices, we'll also explore using 3D CNNs to analyze the images in a more
holistic way, considering the spatial relationships between the slices.

For training the model, we’ll use a loss function that balances out the fact that tumor regions are much
smaller than the rest of the brain, combining Dice Loss and Binary Cross-Entropy. The ultimate goal
is to create a tool that can help doctors make quicker, more accurate diagnoses, potentially leading to
better treatment outcomes for patients with brain tumors.
 REFERENCES

[1] D. Daimary, M. B. Bora, K. Amitabh, D. Kandar, Brain Tumor Segmentation from MRI Images
Hybrid Convolutional Neural Networks, 2020.

[2] B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy-Cramer, K. Farahani, J. Kirby, et al. ”The

Multimodal Brain Tumor Image Segmentation Benchmark (BRATS)”, IEEE Transactions on
Medical Imaging, 2015.

[3] S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J. Kirby, et al., ”Segmentation Labels
and Radiomic Features for the Pre-operative Scans BIBLIOGRAPHY 74 of the TCGA-GBM
collection”, The Cancer Imaging Archive, 2017.

[4] M. Kuhn and K. Johnson Applied Predictive Modelling Volume 615 (2013)

[5] http://braintumorsegmentation.org/

[6] B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy-Cramer, K. Farahani, J. Kirby,”The Multimodal

Brain Tumor Image Segmentation Benchmark (BRATS)”, IEEE Transactions on Medical Imaging
34(10), 1993-2024 (2015) DOI: 10.1109/TMI.2014.2377694

[7] S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J.S. Kirby, et al., ”Advancing The Cancer
Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features”,
Nature Scientific Data, 4:170117 (2017) DOI:10.1038/sdata.2017.117

[8] S. Bakas, M. Reyes, A. Jakab, S. Bauer, M. Rempfler, A. Crimi, et al., ”Identifying the Best
Machine Learning Algorithms for Brain Tumor Segmentation, Progres-
sion Assessment, and Overall Survival Prediction in the BRATS Challenge”, arXivpreprint
arXiv:1811.02629 (2018)