This project was enhanced and submitted as part of the Frostbyte Hackathon 2026, a global competition focused on Artificial Intelligence, Machine Learning, and real-world technological innovation.
DeepX-ChestXray is an AI-powered system designed to assist in the analysis of chest X-ray images using deep learning models and explainable AI techniques.
The system allows users to upload chest X-ray images and receive:
• AI-based classification results
• Probability scores for predictions
• Explainable visualizations using GradCAM++
• Real-time inference through an interactive web interface
This project demonstrates how machine learning and explainable AI can assist medical imaging research and education.
Artificial Intelligence & Machine Learning
This project demonstrates the application of deep learning and explainable AI techniques for medical image analysis, aligning with the AI/ML theme of the Frostbyte Hackathon.
DeepX-ChestXray is an AI-powered web application designed for automated analysis of chest X-ray images. Using state-of-the-art convolutional neural networks (CNNs), the system performs classification of chest X-rays and provides explainable visualizations to help understand model predictions.
The platform is built with an interactive Streamlit interface, making it accessible for researchers, educators, and AI enthusiasts to explore AI-based medical imaging.
You can access the live application here: DeepX-ChestXray Web App
- Classification of chest X-rays into normal or abnormal/pathology-specific categories
- Explainable AI outputs using techniques such as GradCAM++ and feature attribution methods
- Real-time inference for immediate results
- Modular architecture allowing integration of new models or additional pathologies
- Visualization overlays highlighting regions of interest for interpretability
DeepX-ChestXray/
├── app.py # Main Streamlit application script
├── models/ # Pre-trained CNN models
│ ├── cnn_model_final.keras
│ └── cnn_model_backup.keras
├── assets/ # Static frontend resources (styles, icons)
├── requirements.txt # Python dependencies
├── README.md # Project documentation
└── data/ # Optional folder for datasets
- Clone the repository:
git clone https://github.com/yourusername/DeepX-ChestXray.git
cd DeepX-ChestXray- Install dependencies:
pip install -r requirements.txt
- Run the application:
streamlit run app.py
Chest X-Ray Pneumonia Dataset from Kaggle:
https://www.kaggle.com/datasets/paultimothymooney/chest-xray-pneumonia
The dataset is organized into train/, val/, and test/ folders with subfolders:
NORMAL— images of healthy patientsPNEUMONIA— images of patients diagnosed with pneumonia
Upload a chest X-ray image (JPEG/PNG). The app will display:
- Predicted classification (e.g., Normal, Pneumonia, TB)
- Explainability visualization (heatmap or GradCAM++ overlay)
- Probability scores for model predictions
- Model Architecture: Utilizes a CNN backbone (e.g., ResNet, DenseNet, or custom CNN) trained on chest X-ray datasets.
- Preprocessing: Input images are resized, normalized, and optionally converted from DICOM format.
- Explainability: GradCAM++ and feature attribution methods (like SHAP or Integrated Gradients) generate visual explanations of model focus areas.
- Inference: Real-time predictions are provided via Streamlit interface.
- Deployment: The web app allows interactive uploading, inference, and visualization without requiring code execution by the user.
- Research and prototyping of AI in medical imaging
- Educational purposes to demonstrate AI interpretability
- Not intended for clinical diagnosis; all results should be validated by certified radiologists
- Performance depends heavily on training data diversity and quality
- Explainability overlays show areas the model focused on, not definitive diagnosis
- Privacy and regulatory compliance are critical when handling medical images; this app is for research/demo only
- Multi-class pathology detection (e.g., pneumonia, tuberculosis, lung opacity)
- Support for DICOM medical imaging format
- External validation on independent datasets for generalization
- Confidence scoring / uncertainty estimation for predictions
- Structured report export in PDF or JSON summarizing predictions and visual explanations
- Selvaraju, R.R., Cogswell, M., Das, A., et al., Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization, ICCV 2017.
- Lundberg, S.M., Lee, S.I., A Unified Approach to Interpreting Model Predictions, NIPS 2017.