WCLC 2025 - Posters & ePosters-P1.04.53 Potential for AI as First Reader in Lung Cancer Screening

Pdf Summary

This study evaluated the agreement between human radiologists and artificial intelligence (AI) software in interpreting low-dose CT (LDCT) scans for lung cancer screening (LCS) using the Lung Reporting and Data System (LungRADS) categories. LungRADS classifies scans as negative (categories 1-2) or positive (categories 3-4) for lung cancer risk. The retrospective analysis involved 4,104 baseline LDCT scans from volunteers enrolled in the BioMILD trial. Experienced radiologists prospectively assigned LungRADS scores, which were then retrospectively compared to AI-generated LungRADS categories. Lung cancer diagnoses within two years post-baseline served as the reference standard.<br /><br />The agreement between AI and human readings was measured using the weighted kappa (k-Cohen Index with Fleiss-Cohen weights), indicating moderate concordance. The results showed that AI readings had comparable diagnostic accuracy to human readings for detecting lung cancer. Notably, utilizing AI as a first reader could potentially reduce the workload of human radiologists by approximately 74.7%, as many LDCT scans could be initially evaluated by AI and only those flagged as positive or uncertain would require human review.<br /><br />The study concludes that AI can safely serve as a first reader in lung cancer screening programs, potentially streamlining the screening process without compromising diagnostic reliability. This finding supports further research to optimize AI integration into clinical workflows, aiming to enhance efficiency and accuracy in lung cancer detection.<br /><br />The research was conducted by a multidisciplinary team from Italian institutions, including Fondazione IRCCS Istituto Nazionale dei Tumori and universities of Parma and Turin. The study is registered under ClinicalTrials.gov Identifier: NCT02247453.

Asset Subtitle

Giovanni Leuzzi

Meta Tag

Speaker Giovanni Leuzzi

Topic Screening and Early Detection

Keywords

lung cancer screening

low-dose CT

LungRADS

artificial intelligence

radiologist agreement

BioMILD trial

diagnostic accuracy

weighted kappa

AI first reader

clinical workflow integration