2026 IASLC CT Screening and Early Detection Symposium-WS02.17 - Talk 3: How AI Can Streamline Lung Cancer Screening Programs

Pdf Summary

The presentation explains how AI can streamline lung cancer screening (LCS) programs using low-dose CT (LDCT). It covers the full workflow: eligibility assessment, CT acquisition and reconstruction, interpretation, reporting, and risk estimation. AI can help improve image quality through noise reduction, spatial-resolution enhancement, and kernel conversion, then support detection and characterization of nodules, including solid, part-solid, and ground-glass nodules. It can also measure nodule size, volume, growth, and doubling time, and estimate malignancy risk and future cancer risk.<br /><br />Beyond nodules, AI can quantify important non-nodule findings such as coronary artery calcium, emphysema/airway disease, interstitial lung abnormalities, thoracic aortic aneurysm, and body composition. The talk also highlights different reading modes: second reader, concurrent reader, first reader, and backup/standalone reader. Because screening populations have low disease prevalence, AI needs especially high sensitivity to avoid false negatives.<br /><br />A key point is that in low-prevalence settings, even excellent sensitivity and specificity may still yield low positive predictive value, making workflow design crucial. Evidence from recent studies suggests AI can be effective as a prescreener or first reader. In the 4-IN-THE-LUNG-RUN trial, AI had far fewer negative misclassifications and missed referrals than radiologists. In the BioMILD study, AI achieved comparable sensitivity with lower specificity and could reduce radiologist workload substantially.<br /><br />The presentation also describes a cloud-based, networked LCS system with central and regional quality-control centers, supporting image upload, reading systems, education, monitoring, and nodule management. Finally, it notes that retrospective central reading may reduce inter-institution variability compared with prospective institutional reading, suggesting AI and centralized systems could improve consistency and efficiency in LCS programs.

Asset Subtitle

Jin Mo Goo

Meta Tag

Concept Chest Screening

Concept Lung Cancer Screening CT

Concept Nodule detection

Concept Malignancy prediction

Concept Nodule Classification

Speaker Jin Mo Goo

Keywords

lung cancer screening

low-dose CT

AI workflow

nodule detection

image quality enhancement

malignancy risk estimation

coronary artery calcium

prescreener

centralized reading

false negatives

Chest Screening

Lung Cancer Screening CT

Nodule detection

Malignancy prediction

Nodule Classification

Jin Mo Goo