Related Experiment Video
Updated: Sep 12, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
External Testing of a Deep Learning Model for Lung Cancer Risk from Low-Dose Chest CT
Jong Hyuk Lee1,2, Kum Ju Chae3, Michael T Lu4,5
1Department of Radiology, Seoul National University Hospital, 101, Daehak-ro, Jongno-gu, Seoul 03080, Korea.
Radiology
|August 5, 2025
Summary
Sybil, a deep learning model for lung cancer prediction, showed strong performance in Asian individuals with heavy smoking histories but struggled with those who never smoked or smoked lightly, especially for future cancer detection.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Thoracic Oncology
- Radiomics and Deep Learning
Background:
- Sybil, an open-source deep learning model, requires external validation for lung cancer prediction using low-dose CT (LDCT).
- Its effectiveness in identifying high-risk individuals who are never-smokers or light-smokers is not well-established.
Purpose of the Study:
- To externally validate Sybil's performance in predicting lung cancer risk in an Asian health checkup cohort.
- To assess Sybil's utility in identifying high-risk individuals, including never-smokers and light-smokers.
Main Methods:
- Retrospective analysis of LDCT scans from 18,057 individuals (aged 50-80) with follow-up data.
- Evaluation of Sybil's predictive performance using time-dependent AUC for 1-year and 6-year lung cancer risk.
- Subgroup analysis based on smoking history (heavy vs. never/light) and lung cancer visibility on baseline scans.
Main Results:
- Sybil achieved high AUCs (0.91 for 1-year, 0.74 for 6-year) in the overall cohort.
- In heavy smokers, Sybil demonstrated strong performance for visible lung cancers (AUC 0.94) and acceptable for future cancers (AUC 0.70).
- In never/light smokers, Sybil performed well for visible cancers (AUC 0.89) but poorly for future cancers (AUC 0.56).
Conclusions:
- Sybil shows excellent performance for visible lung cancers and acceptable for future lung cancers in Asian heavy smokers.
- Sybil's performance for detecting future lung cancers is poor in Asian individuals with never or light smoking histories.
