Performance and reading time of lung nodule identification on multidetector CT with or without an artificial...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Lung nodules are a critical finding in computed tomography (CT) scans, requiring accurate detection for early diagnosis.
• Computer-aided detection (CAD) systems utilizing artificial intelligence (AI) offer potential improvements in radiologist performance.

Purpose of the Study:

• To compare the diagnostic performance and reading efficiency of radiologists using AI-powered CAD for lung nodule detection in different reading modes.
• To evaluate the impact of CAD on both junior and senior readers.

Main Methods:

• Retrospective analysis of 150 multidetector CT datasets with 340 lung nodules (≤10 mm).
• Readers (3 junior, 3 senior) interpreted images in three modes: without CAD, second-read with CAD, and concurrent read with CAD.
• Diagnostic performance (sensitivity, specificity, AUC) and reading times were compared.

Main Results:

• Mean sensitivity significantly improved with CAD (64% without CAD to 82% in second-read and 80% in concurrent-read modes).
• No significant difference in sensitivity, specificity, or AUC between second-read and concurrent-read modes.
• Concurrent-reading mode significantly reduced reading time (124s) compared to without CAD (156s) and second-read (197s).

Conclusions:

• AI-powered CAD enhances lung nodule detection performance for all readers.
• Both second-read and concurrent-read modes are beneficial for screening and routine practice.
• Concurrent reading with CAD offers greater efficiency for radiologists of all experience levels.