Reliable and Explainable Brain Tumor Segmentation

A robust AI-driven system for reliable and explainable brain tumor segmentation.
Getting Started

View Demo

Table of Contents

1. About The Project
2. Getting Started
   • Folder Structure
   • Prerequisites
   • Installation
3. Usage
4. Contributing
5. Git Contribution Guide
6. Git Contribution Guide
7. Git Contribution Guide
8. License
9. Contact

About The Project

Brain tumor segmentation is a challenging task requiring radiologists to manually analyze MRI scans. This project aims to develop a robust, explainable AI model for brain tumor segmentation, addressing challenges such as:

• Large 3D image sizes and multi-modal MRI data
• Multi-class segmentation complexity
• Data scarcity and privacy concerns
• Model robustness and adversarial training
• Explainability for clinical trustworthiness

Key Techniques:

• Attention-based models for multi-modal MRI segmentation
• Differential privacy for preserving patient confidentiality
• Adversarial training for robustness
• Grad-CAM and attention scores for model explainability

Results

Model: Attention UNet

Metric	Value
Accuracy	0.99405
Prediction	0.893865
Recall	0.895307
F1 Score	0.894494
Dice Score	0.773027

Getting Started

Folder Structure

BrainTumorSegmentation/
│
├── documents/
├── data/
├── notebooks/
├── results/
├── requirements.txt
└── README.md

Prerequisites

Installation

(back to top)

Usage

(back to top)

Git Contribution Guide

# Initialize a new Git repository (if not already initialized)
git init

# Add the remote repository
git remote add origin https://github.com/hakeematyab/Reliable-Explainable-Brain-Tumor-Segmentation.git

# Pull the latest changes from the repository
git pull

# (Based on the need) Create a new branch for your changes
git checkout -b feature-branch

# (OR) Checkout to the main branch
git checkout main

# Add your changes
git add .

# Commit your changes with a descriptive message
git commit -m "Added feature XYZ"

# Push your changes to the remote repository
git push

# (Based on the need) Create a pull request from GitHub UI and request review

Git Setup on SLURM HPC

Configure Git Username & Email

git config --global user.name "Your Name"
git config --global user.email "[email protected]"

Authenticate with GitHub/GitLab on SLURM HPC

Since password authentication is no longer supported for GitHub/GitLab, follow the following method.

SSH Authentication**

Generate an SSH Key

ssh-keygen -t ed25519 -C "[email protected]"

Press Enter for the default location (~/.ssh/id_ed25519) and other prompts.

Add the Key to SSH Agent

eval "$(ssh-agent -s)"
ssh-add ~/.ssh/id_ed25519

Copy and Add the SSH Key to GitHub/GitLab

cat ~/.ssh/id_ed25519.pub

Go to GitHub → Settings → SSH and GPG Keys → New SSH Key. Paste the key and save it.

Test SSH Connection

ssh -T [email protected]

It should return:

Hi <your-username>! You've successfully authenticated, but GitHub does not provide shell access.

Set Remote to Use SSH

git remote set-url origin [email protected]:hakeematyab/Reliable-Explainable-Brain-Tumor-Segmentation.git

Register Environment to Kernel

1. Activate the environment.

   conda activate BTSeg

2. Install jupyter and ipykernel.

   pip install jupyter ipykernel
   # If the correct kernel doesn't load
   pip install --force-reinstall jupyter
   pip install --force-reinstall ipykernel

3. Register the environment to jupyter kernel.

   python -m ipykernel install --user --name=BTSeg --display-name "Python-BTSeg" 

(back to top)

License

(back to top)

Contact

Atyab Hakeem - [email protected]

Marko Krstulovic - [email protected]

Project Link: Reliable-Explainable-Brain-Tumor-Segmentation

(back to top)