Sybil - Lung Cancer Risk Prediction

🎯 Model Description

Sybil is a validated deep learning model that predicts future lung cancer risk from a single low-dose chest CT (LDCT) scan. Published in the Journal of Clinical Oncology, this model can assess cancer risk over a 1-6 year timeframe.

Key Features

Single Scan Analysis: Requires only one LDCT scan
Multi-Year Prediction: Provides risk scores for years 1-6
Validated Performance: Tested across multiple institutions globally
Ensemble Approach: Uses 5 models for robust predictions

🚀 Quick Start

Installation

pip install huggingface-hub torch torchvision pydicom

Basic Usage

from huggingface_hub import snapshot_download
import sys

# Download model
model_path = snapshot_download(repo_id="Lab-Rasool/sybil")
sys.path.append(model_path)

# Import model
from modeling_sybil_wrapper import SybilHFWrapper
from configuration_sybil import SybilConfig

# Initialize
config = SybilConfig()
model = SybilHFWrapper(config)

# Prepare your DICOM files (CT scan slices)
dicom_paths = ["scan1.dcm", "scan2.dcm", ...]  # Replace with actual paths

# Get predictions
output = model(dicom_paths=dicom_paths)
risk_scores = output.risk_scores.numpy()

# Display results
print("Lung Cancer Risk Predictions:")
for i, score in enumerate(risk_scores):
    print(f"Year {i+1}: {score*100:.1f}%")

📊 Example with Demo Data

import requests
import zipfile
from io import BytesIO
import os

# Download demo DICOM files
def get_demo_data():
    cache_dir = os.path.expanduser("~/.sybil_demo")
    demo_dir = os.path.join(cache_dir, "sybil_demo_data")

    if not os.path.exists(demo_dir):
        print("Downloading demo data...")
        url = "https://www.dropbox.com/scl/fi/covbvo6f547kak4em3cjd/sybil_example.zip?rlkey=7a13nhlc9uwga9x7pmtk1cf1c&dl=1"
        response = requests.get(url)

        os.makedirs(cache_dir, exist_ok=True)
        with zipfile.ZipFile(BytesIO(response.content)) as zf:
            zf.extractall(cache_dir)

    # Find DICOM files
    dicom_files = []
    for root, dirs, files in os.walk(cache_dir):
        for file in files:
            if file.endswith('.dcm'):
                dicom_files.append(os.path.join(root, file))

    return sorted(dicom_files)

# Run demo
from huggingface_hub import snapshot_download
import sys

# Load model
model_path = snapshot_download(repo_id="Lab-Rasool/sybil")
sys.path.append(model_path)

from modeling_sybil_wrapper import SybilHFWrapper
from configuration_sybil import SybilConfig

# Initialize and predict
config = SybilConfig()
model = SybilHFWrapper(config)

dicom_files = get_demo_data()
output = model(dicom_paths=dicom_files)

# Show results
for i, score in enumerate(output.risk_scores.numpy()):
    print(f"Year {i+1}: {score*100:.1f}% risk")

Expected output for demo data:

Year 1: 2.2% risk
Year 2: 4.5% risk
Year 3: 7.2% risk
Year 4: 7.9% risk
Year 5: 9.6% risk
Year 6: 13.6% risk

📈 Performance Metrics

Dataset	1-Year AUC	6-Year AUC	Sample Size
NLST Test	0.94	0.86	~15,000
MGH	0.86	0.75	~12,000
CGMH Taiwan	0.94	0.80	~8,000

🏥 Intended Use

Primary Use Cases

Risk stratification in lung cancer screening programs
Research on lung cancer prediction models
Clinical decision support (with appropriate oversight)

Users

Healthcare providers
Medical researchers
Screening program coordinators

Out of Scope

❌ Diagnosis of existing cancer
❌ Use with non-LDCT imaging (X-rays, MRI)
❌ Sole basis for clinical decisions
❌ Use outside medical supervision

📋 Input Requirements

Format: DICOM files from chest CT scan
Type: Low-dose CT (LDCT)
Orientation: Axial view
Order: Anatomically ordered (abdomen → clavicles)
Number of slices: Typically 100-300 slices
Resolution: Automatically handled by model

⚠️ Important Considerations

Medical AI Notice

This model should supplement, not replace, clinical judgment. Always consider:

Complete patient medical history
Additional risk factors (smoking, family history)
Current clinical guidelines
Need for professional medical oversight

Limitations

Optimized for screening population (ages 55-80)
Best performance with LDCT scans
Not validated for pediatric use
Performance may vary with different scanner manufacturers

📚 Citation

If you use this model, please cite the original paper:

@article{mikhael2023sybil,
  title={Sybil: a validated deep learning model to predict future lung cancer risk from a single low-dose chest computed tomography},
  author={Mikhael, Peter G and Wohlwend, Jeremy and Yala, Adam and others},
  journal={Journal of Clinical Oncology},
  volume={41},
  number={12},
  pages={2191--2200},
  year={2023},
  publisher={American Society of Clinical Oncology}
}

🙏 Acknowledgments

This Hugging Face implementation is based on the original work by:

Original Authors: Peter G. Mikhael & Jeremy Wohlwend
Institutions: MIT CSAIL & Massachusetts General Hospital
Original Repository: GitHub
Paper: Journal of Clinical Oncology

📄 License

MIT License - See LICENSE file

🔧 Troubleshooting

Common Issues

Import Error: Make sure to append model path to sys.path
```
sys.path.append(model_path)
```

Missing Dependencies: Install all requirements

pip install torch torchvision pydicom sybil huggingface-hub

DICOM Loading Error: Ensure DICOM files are valid CT scans

import pydicom
dcm = pydicom.dcmread("your_file.dcm")  # Test single file

Memory Issues: Model requires ~4GB GPU memory

import torch
device = 'cuda' if torch.cuda.is_available() else 'cpu'

📬 Support

HF Model Issues: Open issue on this repository
Original Model: GitHub Issues
Medical Questions: Consult healthcare professionals

🔍 Additional Resources

Note: This is a research model. Always consult qualified healthcare professionals for medical decisions.

Downloads last month: 22