Sybil: Predicting Future Lung Cancer Risk From a Single Low-Dose Chest Computed Tomography

Abstract

Low-dose computed tomography (LDCT) for Jung cancer screening is effective, though most eligible people are not being screened. Tools that provide personalized future cancer risk assessment could focus approaches toward those most likely to benefit. We hypothesize that a deep learning model assessing the entire volumetric LDCT data could be built to predict individual risk without requiring additional demographic or clinical data. We develop a model called Sybil using LDCTh from the National Lung Screening 'Trial (NLST). Sybil requires only one LDCT and does not require clinical data or radiologist annotations; it can run in real-time in the background on a radiol­ogy reading station. Sybil is validated on three independent datasets: a held-out set of 6,282 LDCTs from NLST participants, 8,821 LDCTs from Massachusetts General Hospital (MGH) and 12,280 LDCTs from Chang Cung Memorial Hospital (CGMH, which included people with a range of smoking history including non-smokers). Sybil achieves areas under the receiver-operator curve for Jung cancer prediction at 1-year of 0.92 (95% CI 0.88, 0.95) on NLST, 0.86 (95% CI 0.82, 0.90) on MGH and 0.94 (95% CI 0.91, 1.00) on CGMH external validation sets. Concordance indices over six years were 0.75 (95% CI 0.72, 0.78), 0.81 (95% CI 0.77, 0.85), and 0.80 (95% CI 0.75, 0.86) for NLST, MGH, and CGMH, respectively. The model is publicly available at https://github.com/reginabarzilaygroup/Sybil.