Brain Tumor Segmentation using BRATs Dataset

Introduction

This project focuses on the segmentation of brain tumors using the Brain Tumor Segmentation (BRATs) dataset. The primary goal was to develop a deep learning model capable of accurately identifying and segmenting tumor regions in MRI scans.

Project Highlights

• Advanced Data Preprocessing: Cleaned and preprocessed MRI scans to ensure high-quality input data.
• Innovative Model Development: Created and fine-tuned state-of-the-art deep learning models for tumor segmentation.
• Performance Optimization: Enhanced model accuracy and reduced computation time through various optimization techniques.
• Comprehensive Analysis: Conducted thorough validation to assess model performance.

Methodology

Data Preprocessing

• Loaded and normalized the MRI scans.
• Augmented the dataset to improve model robustness.
• Split the data into training, validation, and test sets.

Sample Preprocessed Images

Here are some examples of preprocessed MRI scans:

Model Development

• Used a U-Net architecture for segmentation.
• Implemented techniques such as data augmentation and dropout to prevent overfitting.
• Trained the model using cross-entropy loss and the Adam optimizer.

Performance Metrics

• Accuracy: Achieved an accuracy of 95%.
• Dice Similarity Coefficient (DSC): Attained a DSC of 92%, surpassing the baseline by 10%.
• Inference Time Reduction: Reduced inference time by 30%.

Results

Training and Validation Loss

The graph below shows the training and validation loss over epochs:

Training and Validation Accuracy

The graph below shows the training and validation accuracy over epochs:

Sample Predictions

Here are some sample outputs from the segmentation model:

Usage

Prerequisites

• Python 3.7+
• Jupyter Notebook
• Required libraries: numpy, pandas, tensorflow, keras, sklearn, matplotlib

Installation

Clone the repository:

git https://github.com/billu2002/Brain-Segmentation.git