Bristol Stool Scale Classifier - Computer Vision System

An AI-powered computer vision application that classifies stool samples according to the Bristol Stool Scale using deep learning.

Features

• 📸 Multiple Input Methods: Upload images, take photos with camera, or provide image URLs
• 🤖 Advanced AI Model: Uses ResNet50 architecture for accurate classification
• 🎨 Image Preprocessing: Automatic contrast enhancement and noise reduction
• 📊 Confidence Scores: Real-time probability distribution across all 7 types
• 💡 Personalized Recommendations: Health tips based on classification results
• 📈 Analysis History: Track and export your analysis history
• 🤝 Continuous Learning: Users provide feedback to improve the model
• 🔧 Admin Dashboard: Review and classify user submissions
• 🚀 Model Retraining: Automated workflow for continuous improvement

Quick Start

Run Locally

1. Install dependencies:

pip install -r requirements.txt

2. Run the main app:

streamlit run streamlit_app.py

The app will be available at http://localhost:8501

Deploy to Streamlit Cloud

1. Push your code to GitHub
2. Go to https://streamlit.io/cloud
3. Click "Deploy an app"
4. Enter: waltgarcia/stool-AI and file streamlit_app.py

Your app will be live at: https://blank-app-vdw2s266yel.streamlit.app/

🔄 Training Pipeline

This project includes a workflow to continuously improve the model:

User Feedback Collection

• Users upload images and get predictions
• They can confirm or correct the classification
• Images are automatically saved for training

Admin Review

streamlit run admin_dashboard.py

• Review user submissions
• Manually classify if needed
• Move approved images to training dataset

Model Retraining

python retrain_model.py

• Trains new model with user-provided data
• Automatically saves best weights
• Model improves with more data

See TRAINING_WORKFLOW.md for detailed instructions.

📁 Project Structure

stool-AI/
├── streamlit_app.py           # Main application
├── admin_dashboard.py         # Admin panel for reviewing submissions
├── retrain_model.py          # Script to train improved models
├── train_model.py            # Original training script
│
├── data/
│   ├── user_submissions/     # User uploaded images
│   └── bristol_stool_dataset/ # Training dataset (type_1 through type_7)
│
├── model_weights.pth         # Current model weights
├── requirements.txt          # Python dependencies
└── README.md                 # This file

🏥 Bristol Stool Scale

The Bristol Stool Scale classifies stool into 7 types:

Type	Name	Classification
1	Separate hard lumps	Severe Constipation
2	Lumpy and sausage-like	Mild Constipation
3	Sausage-like with cracks	Normal
4	Smooth and soft sausage	Normal (Ideal)
5	Soft blobs	Normal
6	Mushy consistency	Borderline Diarrhea
7	Liquid consistency	Diarrhea

⚙️ Technologies

• Frontend: Streamlit
• ML Framework: PyTorch
• Computer Vision: torchvision, OpenCV
• Data Processing: Pandas, NumPy
• Visualization: Plotly, Matplotlib

📊 Model Architecture

• Backbone: ResNet50
• Input Size: 224×224 pixels
• Output Classes: 7 (Bristol Stool Types)
• Preprocessing: Normalization + CLAHE enhancement

🚀 Deployment

The app is deployed on Streamlit Community Cloud and accessible at:

https://stool-ai-waltgarcia.streamlit.app

📝 License

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

⚠️ Disclaimer

This tool is for informational and educational purposes only. It is not a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of a physician or qualified health provider with questions about medical conditions.

👤 Author

Walt Garcia - GitHub

🤝 Contributing

Contributions are welcome! You can help by:

1. Using the app and providing feedback
2. Uploading images to improve the model
3. Reporting bugs or suggesting features
4. Sharing the app with others

---

Made with ❤️ using Streamlit and PyTorch