supakornn/Brain-Tumor-Classification

Brain Tumor Classification

CNN-based classification of brain tumor types from MRI scans.

Dataset

Source: Brain Tumor MRI Dataset (Kaggle)

Dataset Structure:

The dataset contains MRI images organized into 4 classes:

• glioma - Glioma brain tumors
• meningioma - Meningioma brain tumors
• notumor - Healthy brain scans (no tumor)
• pituitary - Pituitary brain tumors

Each class has images split into Training/ and Testing/ folders.

Model Architecture

SimpleCNN with 10.7M parameters:

• 3 Convolutional blocks (32, 64, 128 filters)
• Batch Normalization + Dropout
• 2 Dense layers (256, 128 neurons)
• Softmax output (4 classes)

Results

Configuration	Accuracy	Loss	Change
Baseline	42.39%	3.2599	-
Fine-tuned	46.70%	2.8472	+4.31%

Fine-tuned hyperparameters: LR=0.0005, Batch=16, Epochs=40

Per-Class Performance (Fine-tuned)

Class	Precision	Recall	F1-Score
Glioma	1.00	0.14	0.24
Meningioma	1.00	0.21	0.35
No Tumor	0.37	0.92	0.53
Pituitary	0.54	0.66	0.59

Installation

Step 1: Install UV

Install uv if you haven't already:

curl -LsSf https://astral.sh/uv/install.sh | sh

Step 2: Setup Project

# Create virtual environment and install dependencies
uv sync

Step 3: Download Dataset

Use the automated download script:

# Make script executable
chmod +x download_dataset.sh

# Run the script (it will guide you through the setup)
./download_dataset.sh

What the script does:

1. Checks UV installation
2. Sets up virtual environment
3. Installs Kaggle CLI
4. Asks for Kaggle credentials (get from: https://www.kaggle.com/settings → API → Create New Token)
5. Downloads dataset to ./data/
6. Shows statistics

Manual download (alternative):

uv add kaggle
mkdir -p ~/.kaggle
# Create ~/.kaggle/kaggle.json with: {"username":"your_username","key":"your_api_key"}
chmod 600 ~/.kaggle/kaggle.json
uv run kaggle datasets download -d masoudnickparvar/brain-tumor-mri-dataset -p ./data --unzip

Legacy Installation (pip)

pip install -r requirements.txt

Requirements: TensorFlow, NumPy, Pandas, Matplotlib, Seaborn, Scikit-learn, Pillow

Usage

Data Exploration

# With uv
uv run python scripts/data_exploration.py

# Or with activated venv
python scripts/data_exploration.py

Training

# With uv
uv run python scripts/train_models.py

# Or with activated venv
python scripts/train_models.py

Training time: ~2 hours. Saves model to models/SimpleCNN_best.h5.

Prediction

# With uv (example with actual filename)
uv run python scripts/predict_single_image.py data/Testing/glioma/Te-gl_1.jpg

# Or with activated venv
python scripts/predict_single_image.py data/Testing/glioma/Te-gl_1.jpg

# Available folders: glioma, meningioma, notumor, pituitary

Limitations

• Low accuracy (46.70%) insufficient for clinical use
• Poor glioma detection (14% recall)
• Simple architecture inadequate for medical imaging

Additional Resources

• Dataset Source: Brain Tumor MRI Dataset on Kaggle
• Repository: GitHub

License

MIT License - Educational and research purposes.