This study explores the use of AI as a first-read filter in low-dose CT (LDCT) lung cancer screening (LCS) to alleviate the increased workload for radiologists. It involved 1254 baseline LDCT thorax scans from a European lung cancer screening trial. These scans were independently evaluated by three radiologists with different levels of experience, a scientific researcher trained in lung nodule detection and segmentation, and an AI prototype. The largest solid nodule detected by each reader and the AI was further analyzed. Discrepancies were reviewed by a consensus panel of two experienced radiologists.
The study found that 65.2% of cases had no nodules or nodules smaller than 100mm³, while 34.8% had larger nodules. The AI demonstrated fewer negative misclassifications than manual readers and had a negative predictive value of 0.92. The results suggest that radiologists would only need to review 35% of baseline LDCT cases with nodules ≥100mm³ if AI were used as a first-read filter. The study concludes that AI could significantly reduce radiologists' workload by accurately ruling out negative cases. Future analysis will focus on the histological results of detected lung cancers to calculate false-positive and negative rates.