AI Model Sybil Shows Promise in Predicting Lung cancer Risk in Heavy Smokers
Table of Contents
A new deep-learning model,dubbed Sybil,demonstrates significant potential in predicting lung cancer risk,particularly among individuals with a history of heavy smoking,according to research published August 5 in Radiology. The findings suggest a pathway toward more tailored lung cancer screening and follow-up protocols.
The study, led by Jong Hyuk Lee, MD, PhD, of Seoul National University Hospital in South Korea, highlights Sybil’s ability to analyze low-dose CT (LDCT) scans and estimate an individual’s risk.”Sybil… demonstrated potential for future lung cancer prediction among Asian individuals with heavy smoking histories of at least 20 pack-years and may support optimization of follow-up intervals,” the research team noted.
The Rise of Proactive Screening in East Asia
In East Asia, opportunistic LDCT screening is increasingly common, driven by a proactive approach to health among individuals. This trend results in a high volume of LDCT scans being performed, even among those without a significant smoking history. Sybil, an open-source model, leverages this wealth of LDCT data to improve risk assessment. Researchers emphasize that combining the algorithm with specific clinical contexts provides a valuable opportuni
The study revealed varying levels of performance depending on the patient group and the timeframe for risk prediction:
Sybil’s AUC performance for predicting lung cancer risk
| AUC | 1-year Risk | 6-year Risk | |
|---|---|---|---|
| Overall | 0.91 | 0.74 | |
| Heavy-Smoking | 0.94 | 0.70 | |
| Never/Light | 0.89 | 0.56 |
(AUC values range from 0 to 1, with higher values indicating better performance.)
According to the researchers, “Sybil demonstrated excellent discriminative performance for visible lung cancers and acceptable performance for future lung cancers in Asian individuals with heavy smoking history but demonstrated poor performance for future lung cancers in a never- or light-smoking subgroup.”
An example provided in the study illustrates Sybil’s accuracy: an axial LDCT image of a 54-year-old heavy smoker revealed a 1.5-cm subsolid nodule, with the model’s “attention map” highlighting the area of concern. Over two years, the nodule grew to 2 cm and was diagnosed as adenocarcinoma. Sybil’s risk scores progressively increased over time, ranging from 13.1% for 1-year risk to 31.2% for 6-year risk.
The Future of AI in lung Cancer Screening
Ongoing research will contribute to the “growing evidence and validation of this novel tool,” according to Francine Jacobson, MD, and Suzanne Byrne, MD, of Brigham and Women’s Hospital in boston, in an accompanying commentary. They suggest that future meta-analyses could unlock the potential for more nuanced,individualized care.
“Ultimately, meta-analyses should be performed wherein AI may offer more nuanced, individualized care as understanding of as-yet-unkown risk factors and genetics at population and individual levels reveal to new inputs for AI that would not be possible for the radiologist or clinician to provide,” they concluded.
