Sybil: A Validated Deep Learning Model to Predict Future Lung Cancer Risk From a Single Low-Dose Chest Computed Tomography
Problem
Most eligible people are not screened for lung cancer, and tools for personalized future risk from a single LDCT were lacking.
Why it matters
Better risk stratification could focus screening on people most likely to benefit from early detection.
What the AI did
Sybil, a deep learning model, uses only one volumetric LDCT without clinical data or radiologist annotations to predict individual future lung cancer risk in real time.
Results
Validated on three independent datasets: 6,282 NLST heldout, 8,821 MGH, and 12,280 CGMH LDCTs; performance metrics truncated
Impact
A validated risk-prediction tool from a single scan; clinical adoption and outcome impact not stated.
Record
| Venue | Journal of Clinical Oncology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Peter G. Mikhael, Jeremy Wohlwend, Adam Yala, Ludvig Karstens, Justin Xiang, Angelo K. Takigami, Patrick Bourgouin, PuiYee Chan et al. |
| Citations | 285 (OpenAlex, at import time) |
| References | 34 |
| Model or method | Sybil (deep learning model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1200/jco.22.01345 |
| Open access | hybrid |
| Keywords | deep learning, lung cancer, LDCT, risk prediction, screening |
Importance rating
major — millions of lives, an industry or a field
Lung cancer screening gap is large ('most eligible people not screened') and validated on ~27,000 scans across three datasets, but abstract gives no mortality or population figures.