🧠 Brain Tumor Classification Using Deep Learning

📖 Overview

This project focuses on the classification of brain tumors from MRI scans using advanced deep learning techniques. The goal is to assist healthcare professionals in the early detection and diagnosis of brain tumors. The project uses transfer learning with the pre-trained Xception model and a custom-built Convolutional Neural Network (CNN) to classify MRI images into four categories:

• Glioma
• Meningioma
• Pituitary
• No Tumor

An interactive Streamlit web app has been developed to allow users to upload MRI scans and receive predictions along with visual explanations.

📸 Screenshots

1. Streamlit Web App:

---

🌟 Features

🧠 Deep Learning Models

• Xception Model:

  • Pre-trained on the ImageNet dataset and fine-tuned for brain tumor classification.
  • Uses deep convolutional layers for feature extraction and classification.
  • Achieves high generalization and accuracy.

• Custom CNN Model:

  • Built from scratch using Keras.
  • Includes convolutional layers, max pooling, and dense layers.
  • Optimized for MRI image classification, providing robust performance on unseen data.
  • Achieved 98% accuracy on test data.

🌐 Interactive Web App

• Developed using Streamlit to enable real-time MRI scan analysis.
• Users can upload MRI images and get predictions from both models.
• Provides detailed insights and saliency map visualizations for predictions.

🔍 Saliency Map Generation

• Integrated Gemini 1.5 Flash for generating visual explanations of predictions.
• Highlights key regions of MRI scans that influenced the model’s decision.

🏆 Challenges Solved:

• Achieved 99% accuracy using Xception transfer learning and 98% accuracy with the custom CNN model.
• Added multimodal interaction capabilities for deeper insights into the models' decisions.
• Interactive web app allows real-time predictions and visual explanations of the results.

📂 Directory Structure

├── Testing/                # Testing MRI images dataset  
│   ├── glioma/
│   ├── meningioma/
│   ├── notumor/
│   └── pituitary/
├── images/                 # Images for README and documentation
├── drive/                  # Contains pre-trained weights and related assets  
├── saliency_maps/          # Generated visual explanations for MRI scans  
├── sample_data/            # Sample input data for testing the app  
├── app.py                  # Streamlit web app script  
├── cnn_model.h5            # Saved weights for the custom CNN model  
├── xception_model.weights.h5  # Saved weights for the fine-tuned Xception model  
├── [`braintumorclassification.py`](braintumorclassification.py ) # Main script for training and prediction
└── [`README.md`](README.md )               # Project documentation

🧠 Models

1. Xception Model

• Pre-trained on the ImageNet dataset and fine-tuned for brain tumor classification.
• Utilizes deep convolutional layers to handle feature extraction.
• Provides excellent generalization and achieves high accuracy.

2. Custom CNN Model

• Built from scratch using Keras.
• Includes multiple convolutional layers, max pooling, and dense layers.
• Optimized for MRI image classification, achieving robust performance on unseen data.

🚀 How to Run the Project

1. Install Dependencies

Ensure you have Python 3.9+ installed. Then, install the required packages:

pip install -r requirements.txt

2. Run the Streamlit Web App

Launch the app using the following command:

streamlit run app.py

• Upload an MRI image to get predictions from both the Xception and CNN models.

3. Train Models (Optional)

If you'd like to train the models from scratch:

1. Modify the training scripts in the app.py file.
2. Ensure that the Training/ and Testing/ datasets are available.
3. Use TensorFlow/Keras to fine-tune or build the models.

🏆 Challenges Implemented

• High Accuracy: Achieved 98% accuracy with the custom CNN and 99% accuracy with Xception transfer learning.
• Interactive Web App: Enables real-time predictions and saliency map visualizations.
• Multimodal Explanations: Integrated Gemini 1.5 Flash for advanced visual and textual insights into model decisions.
• Comparative Analysis: Dashboard for comparing model predictions.
• Detailed Reports: Generates comprehensive prediction reports for doctors and patients.

🔮 Future Enhancements

• Multimodal LLM Integration: Allow interactive chats with MRI scans to explain predictions in more detail.
• Historical Analysis: Add a feature to compare predictions with historical MRI data.
• Mobile Integration: Create a mobile-friendly app for easier accessibility.

🛠️ Tools & Technologies

• Programming Language: Python
• Deep Learning Framework: TensorFlow, Keras
• Web App Framework: Streamlit
• Visualization: Gemini 1.5 Flash for saliency maps
• Pre-trained Models: Xception

📊 Dataset

The dataset used for this project contains labeled MRI scans categorized into:

• Glioma
• Meningioma
• Pituitary
• No Tumor

Ensure you have appropriate rights to use the dataset for academic and non-commercial purposes.

📈 Results

• Xception Model: 99% accuracy on test data.
• Custom CNN Model: 98% accuracy on test data.
• Real-time Predictions: Achieved within seconds using the Streamlit app.