Can Artificial Intelligence Reliably and Accurately Measure Lower Limb Alignment: A Systematic Review and Meta-Analysis.

Abstract

Lower limb alignment (LLA) measurements are vital for pre-operative assessments and surgical planning in orthopedics. Artificial intelligence (AI) can enhance the precision and consistency of these measurements. This systematic review and meta-analysis evaluates the accuracy and reliability of AI-based approaches in detecting anatomical landmarks and measuring LLA angles, highlighting both their strengths and limitations. Adhering to PRISMA guidelines, we searched PubMed, Scopus, Embase, and Web of Science on July 2024 and included observational studies validating AI-driven LLA measurements. Pooled intraclass correlation coefficients (ICCs) were computed to assess inter-rater reliability between AI and manual measurements. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was used to assess study quality. We reviewed 28 studies with 47,200 patients and 61,253 images; AI demonstrated high reliability in measuring 15 lower limb angles, with pooled ICCs ranging from 0.9811 to 1.0597. Angles like the hip-knee-ankle (HKA; ICC = 0.9987, 95% CI: 0.9975-0.9998) and the mechanical tibiofemoral angle (mTFA; ICC = 1.0001, 95% CI: 1.0001-1.0001) showed near-perfect agreement. In contrast, the joint line convergence angle (JLCA) and femoral anatomical-mechanical angle (FAMA) exhibited lower reliability and significant publication bias. Heterogeneity was substantial across most angles (I² = 63%-100%). These findings highlight the potential of AI for clinical applications while also identifying areas that require refinement and standardization. AI exhibits high reliability and accuracy in measuring key LLA angles, often outperforming manual techniques in both speed and consistency. It holds significant promise as a clinical tool, though challenges with less reliable angles warrant further refinement. Future studies should focus on standardizing landmark definitions and addressing implementation barriers to maximize AI's potential in orthopedic practice.