Lung cancer screening in high-risk never-smokers with artificial intelligence (LC-SHIELD study).

Abstract

8055 Background: Approximately 50% of Asian lung cancer patients are never smokers. Screening with low-dose computed tomography (LDCT) of thorax in high-risk never-smokers with family history may reduce mortality. However, implementation of LDCT screening in Asia faces barriers including high cost and shortage of radiologists. Artificial intelligence (AI) programmes designed for automated detection of lung nodules may serve as first-readers, facilitating cost reduction and improving efficiency. LC-SHIELD is a prospective study designed to evaluate the feasibility and clinical utilization of AI-based lung cancer screening in a high-risk never smoker population. Methods: This study enrolled never smokers, defined as individuals with a lifetime exposure to fewer than 100 cigarettes, aged between 50 and 75 years, with at least one first-degree relative diagnosed with lung cancer. Participants underwent LDCT of thorax, and the scans were analyzed using LungSIGHT, an AI-assisted software fine-tuned with local data for lung nodule detection. Nodules with a maximum diameter of ≥5mm are classified as AI-positive and referred to radiologists for formal reporting and workup. As a gold standard all scans are retrospectively reviewed by radiologists who are blinded to the LungSIGHT results. Primary endpoint is baseline detection rate of early stage lung cancer and secondary endpoints include sensitivity and specificity of LungSIGHT in nodule detection compared to radiologist assessment. The target sample size is 1000 and here we report the interim analysis. Results: Between July and December 2024, total of 405 subjects were enrolled. Median age was 61 (range 50-75) and 266 (66%) were female. Three patients were diagnosed with invasive adenocarcinoma (lung cancer detection rate 0.7%, all EGFR mutation positive) and 12 individuals (3.0%) had suspicious lung nodules requiring further diagnostic workup. Testing of AI algorithm was based on the first 181 subjects. At the testing phase, 78 (43%) were AI-positive with sensitivity and specificity at 81% and 85%, respectively. In the validation cohort (n=224), 86 (39%) were AI-positive with sensitivity and specificity at 73% and 77%, respectively. Conclusions: AI assisted first-reader of screening LDCT in high-risk never smokers is feasible. LungSIGHT showed high sensitivity and specificity in lung nodule detection using standard radiologist assessment as gold standard comparator. Recruitment for the study is ongoing. Clinical trial information: NCT06295497 .