Comparative Performance of 3 Artificial Intelligence Systems for Lung Nodule Characterization in Low-Dose Computed Tomography Screening

This retrospective study evaluates three commercially available AI systems for detecting, characterizing, and classifying lung nodules on low-dose computed tomography (LDCT) scans from 100 subjects. The three platforms — including ClearRead CT (Riverain Technologies) and AVIEW LCS (Coreline Soft) — are anonymized as A1, A2, and A3 in the paper, with no official public disclosure of which label corresponds to which product. Performance was benchmarked against a consensus reference standard established by two thoracic radiologists. Evaluation metrics included sensitivity, specificity, accuracy, Lung-RADS categorization agreement (Cohen Kappa), and intraclass correlation coefficients (ICC) for nodule diameter and volume. Results showed that A2 achieved the best-balanced diagnostic performance (sensitivity 72.2%, specificity 83.4%, accuracy 80.6%) and matched the reference standard almost perfectly for Lung-RADS; A3 showed near-perfect agreement with the reference standard but with lower sensitivity; A1 showed no agreement with the reference standard. The study highlights marked inter-system variability and underscores the importance of understanding each AI system's characteristics before clinical deployment in lung cancer screening workflows.