🧠 NeuroScan - Brain Tumor Classifier

📌 Table of Contents

• Model Overview
• Key Features
• Web Interface
• Installation
  • Manual Setup
  • Docker Setup
• Usage
• Performance Metrics
• Future Work
• Contributing
• License
• Acknowledgments

---

🧩 Model Overview

A powerful CNN model that classifies brain tumors into four categories with 98.32% accuracy:

1. Glioma - Irregular tumors originating from glial cells
2. Meningioma - Well-defined tumors of the meninges
3. Pituitary - Hormone-affecting glandular tumors
4. No-tumor - Healthy brain scans

---

🚀 Key Features

• 98.32% Test Accuracy - State-of-the-art performance
• VGG-Inspired Architecture - Optimized 3x3 kernel design
• Advanced Training:
  • Early stopping & model checkpointing
  • Dynamic learning rate scheduling
  • Real-time data augmentation (rotations/flips)
• Production-Ready:
  • Flask web interface
  • Docker container support
  • REST API endpoints

---

🌐 Web Interface

Features:

• Drag-and-drop MRI upload
• Real-time visualization
• Detailed confidence reports
• Mobile-responsive design

---

📥 Installation

Manual Installation

# Clone repository 
git clone https://github.com/AminRezaeeyan/NeuroScan.git
cd NeuroScan

# Create and activate virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # Linux/Mac
# venv\Scripts\activate  # Windows

# Install dependencies
pip install --upgrade pip
pip install -r requirements.txt

# Run development server
flask run --host=0.0.0.0 --port=5000

Docker Installation

# Clone repository
git clone https://github.com/AminRezaeeyan/NeuroScan.git
cd NeuroScan

# Build and run containers (detached mode)
docker-compose up -d --build

# View logs (optional)
docker-compose logs -f

---

💻 Usage

Programmatic API

The pre-trained model (best_model.keras) is available in the models/ directory and can be integrated into your applications:

from PIL import Image
import numpy as np
from tensorflow.keras.models import load_model

# Load the pre-trained model
model = load_model('models/best_model.keras')

def predict_tumor(image_path):
    """
    Predicts tumor class from MRI image
    Args:
        image_path: Path to MRI image (JPEG/PNG)
    Returns:
        dict: {'class': 'glioma/meningioma/pituitary/notumor', 
              'confidence': float,
              'probabilities': dict}
    """
    # Load and preprocess image using Pillow
    img = Image.open(image_path)
    img = img.resize((224, 224))  # Resize to 224x224
    img = img.convert('RGB')  # Ensure RGB format
    img_array = np.expand_dims(img_array, axis=0)  # Add batch dimension
    
    # Make prediction
    pred = model.predict(img_array)
    classes = ['Glioma', 'Meningioma', 'No Tumor', 'Pituitary']
    
    return {
        'class': classes[np.argmax(pred)],
        'confidence': float(np.max(pred)),
        'probabilities': {cls: float(prob) for cls, prob in zip(classes, pred[0])}
    }

# Example usage
result = predict_tumor("path/to/mri.jpg")
print(result)

Sample Output:

{
    'class': 'Meningioma',
    'confidence': 0.9743,
    'probabilities': {
        'Glioma': 0.0121,
        'Meningioma': 0.9743,
        'Pituitary': 0.0087,
        'No Tumor': 0.0049
    }
}

⚠️ Important Note:
The model automatically normalizes input images in its first layer. Do not manually normalize (divide by 255) as this will cause incorrect predictions. Simply pass the raw image array (0-255 values).

---

📊 Performance Metrics

Metric	Score
Accuracy	98.32%
Precision	98.32%
Recall	98.32%
F1 Score	98.32%

Class-wise Breakdown

Class	Precision	Recall	F1 Score
Glioma	98.99%	98.33%	98.66%
Meningioma	97.67%	96.08%	96.87%
Pituitary	98.66%	98.33%	98.50%
No-tumor	98.06%	100%	99.02%

---

🔮 Future Work

• Continuous Accuracy Improvements: Currently targeting 99%+ through:

  • Advanced attention mechanisms
  • Transformer-based hybrid architectures
  • Improved data augmentation pipelines

• Tumor Segmentation: Developing pixel-level detection

• Clinical Integration:

  • DICOM/PACS support
  • HL7/FHIR compatibility
  • Multi-modal analysis (MRI + CT)

• Edge Deployment:

  • ONNX runtime optimization

---

🤝 Contributions

How to Contribute:

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Priority Areas:

• Improved model interpretability
• Additional medical imaging formats (DICOM, NIfTI)
• Performance optimizations
• UI/UX enhancements
• Documentation improvements

---

📜 License

Distributed under the MIT License. See LICENSE for more information.