Bone Scan Analyzer

An AI-powered tool for analyzing whole-body bone scans to detect metastasis.

Setup

1. Create a virtual environment:

python -m venv venv
source venv/bin/activate

2. Install dependencies:

pip install -r requirements.txt

3. Create required directories:

mkdir -p data/images data/models

4. Place your trained models in the models directory:

• ResNet34 models for each region
• Random forest classifier

5. Copy your bone scan dataset to data/bs-80k/temp

Running the Application

1. Start the backend API:

uvicorn src.backend.main:app --reload

2. Start the Streamlit frontend:

streamlit run src.frontend.app.py

3. Open your browser and navigate to http://localhost:8501

Project Structure

• src/backend: FastAPI backend service
• src/frontend: Streamlit web interface
• models: Trained model files
• data: Dataset directory
• tests: Test files

Model Performance

• Accuracy: 88.10%
• Sensitivity: 73.71%
• Specificity: 96.34%
• F1 Score: 81.85%
• AUC: 89.13%