A machine learning-powered diagnostic tool developed using TensorFlow/Keras, OpenCV, and Streamlit, designed to detect pneumonia in chest X-ray images. The system offers explainability features, clinical recommendations, and report generation, supporting fast and accurate diagnosis in low-resource healthcare settings.

• Overview
• Features
• Technologies Used
• Installation and Setup
• Dataset and Preprocessing
• Usage
• Project Structure
• Contributing
• Authors and Acknowledgements

The Chest X-ray Pneumonia Detection System aims to:

• Provide an accessible, ML-based method to classify chest X-rays as Normal or Pneumonia.
• Integrate a trained Convolutional Neural Network (CNN) with a Streamlit frontend for seamless user interaction.
• Offer clinical insights such as saliency maps, severity grading, and follow-up recommendations.
• Support remote diagnosis by exporting results and advice as a PDF report using the fpdf library.

• Model Inference

  • Classifies chest X-rays into Normal or Pneumonia.
  • Confidence scores and model accuracy displayed.

• Clinical Decision Support

  • Patient symptom descriptions and follow-up recommendations.
  • Severity grading and feature markers (e.g., opacity, consolidation).

• Explainability

  • Saliency maps to visualize attention regions in the image.
  • Expandable model architecture and layer information.

• Report Generation

  • PDF report download with diagnosis summary and advice.

• Performance

  • CPU inference time < 2 seconds per image.
  • Tested on 5,800+ samples with accuracy > 92%.

• Frontend: Streamlit, HTML/CSS (Bootstrap-inspired)
• Backend: Python, TensorFlow/Keras, OpenCV, PIL
• Reporting: fpdf (PDF generation)
• Modeling: CNN, XGBoost (optional comparative mode)
• Database: SQLite (for logging/testing)

1. Clone the Repository:

   git clone https://github.com/Avjot-Chawla/Chest-Xray-Pneumonia-Detection.git
   cd Chest-Xray-Pneumonia-Detection

2. Install Requirements:

   pip install -r requirements.txt

3. Run the Application:

   streamlit run app.py

⚠️ If using a pretrained model, place the .h5 model file in the models/ directory and update the model loading code accordingly.

• Source: Kaggle - Chest X-ray Images (Pneumonia)

• Split: 70% Train, 15% Validation, 15% Test

• Preprocessing:

  • Resize to 180x180 pixels
  • CLAHE contrast enhancement
  • Normalization to [0,1] pixel range
  • Augmentation: random flip, zoom, rotation

1. Upload an Image:

   • Supported formats: .jpg, .jpeg, .png

2. Review Results:

   • View prediction results for both CNN and XGBoost models (if enabled).
   • View saliency maps and architecture details.

3. Download Report:

   • Click “Download Full Report” to save PDF containing analysis and recommendations.

Chest-Xray-Pneumonia-Detection/
├── models/                 # Trained model files
├── app.py                 # Streamlit main app
├── utils/                 # Preprocessing, prediction, and PDF functions
├── assets/                # Sample images or UI resources
├── requirements.txt       # Required Python libraries
└── README.md              # This file

Contributions are welcome! If you'd like to contribute a new feature or fix a bug:

1. Fork the repo.
2. Create a new branch.
3. Submit a pull request with clear documentation.

• Tanushree Borase – RA2211003010575
• Avjot Singh Chawla – RA2211003010584
• Project Supervisor: Dr. Thamizhamuthu R Department of Computing Technologies, SRM Institute of Science and Technology