example.py

"""
Simple example of using Sybil for lung cancer risk prediction
"""

import sys
import os

# Install required packages if needed
# !pip install torch torchvision transformers pydicom torchio sybil

# Import the model
from modeling_sybil_wrapper import SybilHFWrapper
from configuration_sybil import SybilConfig

def predict_cancer_risk(dicom_paths):
    """
    Predict lung cancer risk from DICOM files.

    Args:
        dicom_paths: List of paths to DICOM files from a CT scan

    Returns:
        Risk scores for years 1-6
    """
    # Load model
    print("Loading Sybil model...")
    config = SybilConfig()
    model = SybilHFWrapper.from_pretrained("Lab-Rasool/sybil")

    # Run prediction
    print(f"Processing {len(dicom_paths)} DICOM files...")
    output = model(dicom_paths=dicom_paths, return_attentions=False)

    # Extract risk scores
    risk_scores = output.risk_scores.numpy()

    return risk_scores


def main():
    # Example usage with demo data
    # In practice, replace with actual DICOM file paths
    demo_dicom_paths = [
        "path/to/slice001.dcm",
        "path/to/slice002.dcm",
        # ... more slices
    ]

    # For testing, you can download demo data:
    # Download from: https://github.com/reginabarzilaygroup/Sybil

    print("=" * 50)
    print("Sybil Lung Cancer Risk Prediction")
    print("=" * 50)

    # Uncomment when you have actual DICOM files:
    # risk_scores = predict_cancer_risk(demo_dicom_paths)

    # Print results
    # print("\nLung Cancer Risk Predictions:")
    # print("-" * 30)
    # for year, score in enumerate(risk_scores, 1):
    #     risk_pct = score * 100
    #     print(f"Year {year}: {risk_pct:.2f}% risk")

    print("\nNote: This example requires actual DICOM files.")
    print("Please provide paths to LDCT scan DICOM files.")
    print("\nFor more information:")
    print("- Original paper: https://doi.org/10.1200/JCO.22.01345")
    print("- GitHub: https://github.com/reginabarzilaygroup/Sybil")


if __name__ == "__main__":
    main()