Skip to main content
ACR 2026 Annual Meeting Posters Main banner
Rate ePoster

Informatics and Data Science

AI-assisted decision support and Radiologist efficiency in CT lung cancer screening

Location: Concourse Foyer Kiosk

    First Author(s)

  • MM

    Midhhath Afza Munavar Ali, MD

    Diagnostic Radiology Resident
    Corewell Health, Michigan Sate University
    Farmington Hills, Michigan, United States

    • Co-Author(s)

  • AM

    Andrew Moriarity, MD, MBA

    Executive Vice President, Advanced Radiology Services and Associate Professor
    Advanced Radiology Services and Division of Radiology and Biomedical Imaging, College of Human Medicine, Michigan State University, United States

  • JC

    Jacob Charron, n/a

    Medical Student
    Michigan State University, United States

Purpose: To assess the impact of AI-assisted decision support on radiologist efficiency and dictation variability in CT lung cancer screening using a pre and post implementation analysis.

Methods/Materials: A retrospective pre–post analysis evaluated radiologists’ dictation performance before and after deployment of a commercially available AI-based CT lung screening decision support tool at an urban academic health system. Examinations from the pre-implementation phase; November 2024–March 2025 were compared with those from the post-implementation phase; May–July 2025.
Primary outcomes were percentage change in median dictation time and month-to-month variability. Secondary analyses assessed associations between exam volume, AI training completion, agreement with AI output, and observed efficiency changes. Rolling-average trends assessed temporal stabilization.

Results: Following AI implementation, radiologists demonstrated percentage reductions in median dictation time, ranging from approximately 9% to over 70%, depending on baseline performance and post-implementation exam volume. Higher post-implementation volumes were associated with greater percentage efficiency gains, while minimal exposure showed less change. Month-to-month variability decreased after implementation, reflecting improved consistency.
Disagreement with AI output was associated with increased dictation time due to additional documentation steps. Radiologists who completed formal AI training demonstrated higher agreement rates (>90%) and greater percentage efficiency improvements compared with untrained peers.
Overall agreement across 1,940 examinations over 12 weeks was 88%, with an upward trend and discrepancies that were not clinically consequential.

Conclusions: AI-assisted decision support for CT lung cancer screening was associated with meaningful percentage-based efficiency improvements and reduced performance variability, particularly among trained, high-volume users. Efficiency gains appear driven by workflow adaptation and training adherence rather than absolute time reduction alone, supporting AI integration as a scalable tool for improving consistency and operational performance in population-based lung cancer screening.