Assessing the Reliability of artificial Intelligence-based Long-leg Alignment Analysis for Lower Limb Deformities Using EOS Imaging System: A Pilot Study

Abstract Objectives: The purpose of this pilot study was to evaluate the reliability of artificial intelligence (AI) software in measuring lower limb alignment parameters using EOS imaging in an orthopedic setting. Methods: Twenty EOS long-leg images were included, representing 40 legs, from 10 patients with varying degrees of lower limb deformities. Two readers performed manual measurements twice. The same dataset was also analyzed using a commercially available AI-based software, leg angle measurement assistant. The agreement between manual and AI measurements was assessed across nine parameters, including 3 length and 6 angular measurements. Intraclass correlation coefficients (ICCs) assessed inter- and intrarater reliability, and Bland–Altman plots evaluated bias. Results: The AI system failed to generate results in 25% of cases. In the AI-manual comparison, reliability varied across parameters. Angular measurements showed poor-to-moderate agreement, with ICC values ranging from 0.30 to 0.70. Length measurements, however, demonstrated excellent agreement, with ICC values ranging from 0.97 to 0.99. The reliability and agreement between manual measurements ranged from good to excellent, both within and between readers. For AI analysis, repeated measurements showed an intraobserver ICC of 1. Conclusion: The AI tool showed potential as an aid for long-leg alignment analysis when applied to EOS images, particularly for length measurements, though it struggled with angular measurements. This may stem from differences between EOS and traditional radiographs affecting accuracy. Further refinement and EOS-specific training are needed, especially for cases with implants and severe deformities.