LungAI: An AI Biomarker that Refines Lung-RADS v2022 Triage of Screen-Detected Pulmonary Nodules

LungAI, a deep learning malignancy classifier for screen-detected pulmonary nodules on NLST low-dose CT, achieved AUROC 0.92 (95% CI 0.88–0.95) and improved risk triage beyond size-based features.

• Why it matters: Lung-RADS v2022 relies mainly on nodule size and attenuation, which can misallocate patients between surveillance and invasive workup. Adding imaging-risk information independent of those features could reduce unnecessary procedures while preserving cancer detection.
• What they did: LungAI was validated on the held-out NLST test partition (2,747 examinations; 1,244 patients; 89 cancers; 3.2% prevalence) and compared with Sybil, mean diameter, and estimated volume. A radiologist assigned Lung-RADS v2022 categories to a positives-enriched subset (226 eligible; 78 cancers), then a reclassification algorithm projected shifts across the full cohort using contingency weighting and score cutpoints.
• The result: LungAI discrimination exceeded Sybil, diameter, and volume (e.g., AUROC delta 0.07 vs Sybil; AUPRC delta 0.18), and the combined reclassification reduced benign biopsy referrals by 20% (32.7 → 26.2 per 1,000) with similar total biopsy volume (51.0 → 50.1). With full follow-up adherence, it estimated 25 months net diagnostic lead time, ~$2,200 net cost savings, and 6.5 fewer benign biopsies per 1,000; with published adherence, it yielded 1.6 additional cancer diagnoses within 1

AI summary of the abstract · may contain errors · not medical advice · read the paper.