A deep learning model developed by the Massachusetts Institute of Technology (MIT) and General Hospital, known as Sybil, has demonstrated remarkable accuracy in predicting lung cancer risk from a single low-dose CT (LDCT) scan. The study was conducted in South Korea, AI-based prediction model achieved 86% accuracy for predicting cancer within one year and 74% accuracy within six years, even among individuals with pre-existing lung conditions.
86% accuracy for 1-year prediction from a single low-dose CT (LDCT) scan
The study was published in the American Journal of Respiratory and Critical Care Medicine, and researchers analyzed LDCT scans from 21,087 people aged 50–80, who underwent self-initiated LDCT screening between January 2009 and December 2021 at a tertiary hospital-affiliated screening center. Follow-up continued until June 2024. The findings highlight the tool’s potential in addressing Asia’s rising lung cancer rates, particularly among non-smokers—a group increasingly recognized as high-risk but often overlooked in current screening guidelines.
Sybil’s predictive capability offers a precision approach to lung cancer screening (LCS). By identifying individuals at lower risk, it could help extend screening intervals or even discontinue unnecessary screenings. Conversely, it could flag at-risk individuals who do not currently meet eligibility criteria for standard screening programs. This dual capacity makes Sybil a valuable asset in optimizing healthcare resources, ensuring screening is both targeted and efficient. AstraZeneca, a founding partner of the Lung Ambition Alliance, has supported initiatives that align with this goal, aiming to transform lung cancer outcomes through early detection and tailored interventions.
To read the full abstract, access via: https://www.atsjournals.org/doi/abs/10.1164/ajrccm.2025.211.Abstracts.A501
