🧠 Brain Tumor Detection and Classification

Vision Transformer (ViT) based deep learning project for brain tumor detection and classification from MRI images.

📋 Table of Contents

• About the Project
• Features
• Dataset
• Installation
• Usage
• Model Architecture
• Results
• Project Structure

🎯 About the Project

This project is developed for tumor detection and classification from MRI brain images. Using Vision Transformer (ViT) model, it performs high-accuracy classification between 4 different classes:

• Glioma - Tumor originating from glial cells
• Meningioma - Tumor originating from meninges membrane
• Pituitary - Pituitary gland tumor
• Healthy - Healthy brain

✨ Features

• 🔬 Image Preprocessing Pipeline

  • Automatic black border cropping
  • Noise reduction with bilateral filter
  • CLAHE contrast enhancement
  • Standard resizing (224x224)

• 🤖 Vision Transformer Model

  • Pre-trained ViT-Small model
  • Transfer learning
  • Data augmentation

• 📊 Comprehensive Evaluation

  • Classification metrics (Precision, Recall, F1-Score)
  • Confusion matrix
  • ROC curve and AUC scores

📁 Dataset

📥 Download Dataset

The raw (unprocessed) dataset used in this project is shared on Kaggle. You can easily download it from the link below:

🔗 Brain Tumor MRI Images Classification Dataset

💡 Note: This dataset is created by combining 3 different Kaggle datasets.

📊 Dataset Statistics

Class	Image Count	Ratio
Glioma	3,768	24.1%
Healthy	3,990	25.6%
Meningioma	3,806	24.4%
Pituitary	4,041	25.9%
Total	15,605	100%

🚀 Installation

Requirements

• Python 3.10+
• CUDA enabled GPU (recommended)

Step 1: Clone the Repository

git clone https://github.com/ZehraKucuker/brain_tumor_classification.git
cd brain_tumor_classification

Step 2: Create Virtual Environment

python -m venv .venv

# Windows
.venv\Scripts\activate

# Linux/Mac
source .venv/bin/activate

Step 3: Install Dependencies

pip install -r requirements.txt

Step 4: Install PyTorch with CUDA (for GPU)

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install timm

💻 Usage

Jupyter Notebook

All project code is consolidated into a single interactive notebook file:

jupyter notebook brain_tumor_classification.ipynb

Or you can open it directly in VS Code.

Notebook Content

1. Dataset Analysis

• Number of images in each class
• Image dimensions and statistics
• Pixel value distributions
• Class distribution charts

2. Image Preprocessing

• Black border cropping
• Bilateral filter (noise reduction)
• CLAHE (contrast enhancement)
• 224x224 resizing

3. Model Training

• Vision Transformer (ViT-Small) model
• Transfer learning training
• Data augmentation

4. Model Evaluation

• Confusion Matrix
• ROC curves
• Classification report

Configuration

Parameters can be adjusted from the CONFIG dictionary in the notebook:

CONFIG = {
    'batch_size': 32,
    'epochs': 15,
    'learning_rate': 1e-4,
    'model_name': 'vit_small_patch16_224',
    ...
}

🏗️ Model Architecture

Vision Transformer (ViT-Small)
├── Patch Embedding (16x16 patches)
├── Transformer Encoder (12 layers)
│   ├── Multi-Head Self-Attention
│   └── MLP Block
├── Classification Head
└── Output: 4 classes

Model Features:

• Total Parameters: 21,667,204
• Patch Size: 16x16
• Input Size: 224x224x3
• Pre-trained: ImageNet-21k

📈 Results

Model Performance

Metric	Value
Test Accuracy	99.87%
Macro F1-Score	0.9987
Weighted F1-Score	0.9987

Class-wise Results

Class	Precision	Recall	F1-Score	ROC-AUC
Glioma	1.0000	1.0000	1.0000	1.0000
Healthy	1.0000	1.0000	1.0000	1.0000
Meningioma	0.9947	1.0000	0.9973	1.0000
Pituitary	1.0000	0.9949	0.9975	1.0000

Visualizations

All graphs (training history, confusion matrix, ROC curves) are displayed inline within the notebook.

📂 Project Structure

brain_tumor_classification/
│
├── dataset/                          # Original dataset
│   ├── glioma/
│   ├── healthy/
│   ├── meningioma/
│   └── pituitary/
│
├── dataset_processed/                # Processed dataset (224x224)
│   ├── glioma/
│   ├── healthy/
│   ├── meningioma/
│   └── pituitary/
│
├── .venv/                            # Python virtual environment
│
├── brain_tumor_classification.ipynb  # Main notebook (all code)
├── requirements.txt                  # Python dependencies
├── best_model.pth                    # Trained model weights
│
└── README.md                         # This file

🔧 Configuration

Main configuration parameters in the notebook:

Parameter	Default	Description
batch_size	32	Mini-batch size
epochs	15	Number of training epochs
learning_rate	1e-4	Learning rate
image_size	224	Input image size
model_name	vit_small_patch16_224	ViT model variant
train_split	0.8	Training set ratio
val_split	0.1	Validation set ratio
test_split	0.1	Test set ratio

📚 Dependencies

numpy
pandas
matplotlib
seaborn
opencv-python
scikit-learn
scikit-image
Pillow
tqdm
torch
torchvision
timm