Brain Tumor Detection using VGG16, DenseNet121, and ResNet50

This repository contains a deep learning-based solution for detecting brain tumors using three state-of-the-art Convolutional Neural Network (CNN) architectures: VGG16, DenseNet121, and ResNet50. The models are trained on a brain tumor dataset, where the task is to classify MRI images as either tumor-positive or tumor-negative.

Key Features

• Multiple Model Architectures: Implements VGG16, DenseNet121, and ResNet50 for comparison of performance across architectures.
• Transfer Learning: Pre-trained weights are used to fine-tune the models for better accuracy with less data.
• Data Augmentation: Implements various data augmentation techniques to enhance model generalization and address data imbalance.
• Imbalanced Dataset Handling: The dataset used is imbalanced, and techniques like class weighting or oversampling can be used to address this issue.
• Performance Metrics: Includes evaluation metrics accuracy to measure model performance.
• Visualization: Uses Grad-CAM for visualizing model attention on MRI scans, providing interpretability to the predictions.

Requirements

• Python 3.x
• TensorFlow / Keras
• NumPy
• Matplotlib
• Scikit-learn

How to Use

1. Clone the repository:

   git clone https://github.com/yourusername/Brain-Tumor-Detection.git
   cd Brain-Tumor-Detection

2. Install the required dependencies:

   pip install -r requirements.txt

3. Run the training script:

   python train_model.py

4. Evaluate the model:

   python evaluate_model.py

5. Visualize model predictions:

   python visualize_results.py

Dataset

The dataset used consists of MRI brain images, categorized into two classes: Tumor and No Tumor. Due to the imbalance in class distribution, appropriate techniques such as class weighting or oversampling have been applied.

Results

The results from each model are compared based on accuracy evaluation metrics. The model's performance can be improved by adjusting hyperparameters or further tuning the architectures.

Future Improvements

• Experiment with other CNN architectures.
• Implement model ensembling to improve accuracy.
• Explore further data augmentation techniques and balancing strategies.

License

This project is licensed under the MIT License - see the LICENSE file for details.