AI-driven radiographic assessment of knee osteoarthritis: Advances in Kellgren–Lawrence grading

Knee osteoarthritis (KOA) is a progressive musculoskeletal disorder and a leading cause of disability worldwide. Radiographic assessment using the Kellgren–Lawrence (KL) grading system remains the clinical standard for evaluating disease severity and serves as the primary labeling framework for automated classification models. This review provides a structured analysis of artificial intelligence–based approaches for KL-driven KOA grading from 2013 to 2025.The study examines publicly available radiographic datasets, classical machine-learning pipelines, convolutional neural networks, ordinal-aware optimization strategies, CNN–Transformer hybrid architectures, and emerging representation-learning paradigms. Quantitative performance trends are synthesized to compare methodological categories, with particular attention to sensitivity patterns and grade-adjacent misclassification. Across studies, robust knee localization, ordinal modeling, and data-centric training strategies consistently demonstrate stronger generalization than architectural complexity alone. Although recent transformer-based models and hybrid frameworks show promise, external validation and clinically grounded evaluation remain limited. This review consolidates methodological lessons, identifies persistent challenges such as KL-1 ambiguity and dataset heterogeneity, and outlines future directions emphasizing generalizability, interpretability, and real-world clinical integration.