Brain Tumor Classification via Deep Learning

Project Overview

This project implements a brain tumor classification system using deep learning techniques. The model utilizes a pre-trained VGG16 architecture with transfer learning to classify brain MRI scans into four categories:

• Glioma tumor
• Meningioma tumor
• No tumor
• Pituitary tumor

Dataset

The dataset consists of approximately 4,800 labeled MRI scans divided into pre-defined training and validation sets. The dataset can be downloaded from Kaggle (link to be added).

Methodology

Data Preprocessing

• Images are resized to a uniform size
• Pixel values are normalized to the range [0, 1]
• File paths are systematically created using os.path.join for platform-independent access

Model Architecture

The model leverages the VGG16 architecture with transfer learning:

• The pre-trained VGG16 model serves as a feature extractor
• Initial layers capturing generic image features are kept frozen
• Final layers are replaced with new Dense layers using softmax activation
• The model is adapted for the four-class classification task

Training Process

• Loss Function: Sparse categorical cross-entropy
• Optimizer: Adam
• Metrics: Accuracy
• Early stopping implemented to prevent overfitting
• TensorBoard visualization utilized for monitoring training

Implementation Details

Requirements

• Python 3.x
• TensorFlow/Keras
• NumPy
• Pandas
• Matplotlib (for visualization)
• scikit-learn (for evaluation metrics)

Code Structure

brain-tumor-classification/
├── data/
│   ├── training/
│   └── validation/
├── models/
├── src/
│   ├── preprocess.py
│   ├── model.py
│   ├── train.py
│   └── evaluate.py
└── README.md

Usage

Setup

# Clone the repository
git clone https://github.com/username/brain-tumor-classification.git
cd brain-tumor-classification

# Install dependencies
pip install -r requirements.txt

Training

python src/train.py

Evaluation

python src/evaluate.py

Results

The model was trained for 12 epochs with a batch size of 32. Performance metrics include:

• Overall accuracy
• Precision, recall, and F1-score for each tumor class
• Test loss to evaluate generalization capability

Future Work

• Explore different hyperparameter configurations
• Test alternative pre-trained models
• Implement more extensive data augmentation techniques
• Investigate model explainability for clinical applications

References

1. Saleh, A., Sukaik, R., & Abu-Naser, S. S. (2020). Brain tumor classification using deep learning.
2. Paul, J. S., Plassard, A. J., Landman, B. A., & Fabbri, D. (2017). Deep learning for brain tumor classification.
3. Ari, A., & Hanbay, D. (2018). Deep learning based brain tumor classification and detection system.
4. Díaz-Pernas, F. J., Martínez-Zarzuela, M., Antón-Rodríguez, M., & González-Ortega, D. (2021). A deep learning approach for brain tumor classification and segmentation using a multiscale convolutional neural network.
5. Tandel, G. S., et al. (2019). A review on a deep learning perspective in brain cancer classification.

Author

Muhammad Tahir Zia
Bachelors Computer Engineering
GIK Institute, Topi, Pakistan
u2021465@giki.edu.pk