AI-based knee osteoarthritis progression prediction: a comprehensive global bibliometric and hotspot evolution analysis (2010–2025)

BACKGROUND: Knee osteoarthritis (OA) is a leading global cause of disability, yet conventional tools lack sensitivity for early detection and precise prognostication. Artificial intelligence (AI) and machine learning (ML) offer powerful means to enhance prediction of knee OA onset and progression. This bibliometric study maps global research trends and thematic evolution rather than evaluating the clinical effectiveness of individual AI tools. OBJECTIVE: This study systematically maps the global research landscape on AI-based knee OA progression prediction from 2010 to November 2025, highlighting key contributors, collaboration networks, methodological trends, and evolving research hotspots. METHODS: A comprehensive bibliometric analysis was performed using Web of Science, Scopus, PubMed, and IEEE Xplore. Embase was not included due to substantial overlap (>90%) with PubMed/MEDLINE. Search terms included "artificial intelligence," "machine learning," "deep learning," "knee osteoarthritis," and '"progression prediction." Following systematic deduplication and dual-reviewer screening (Cohen's κ = 0.89), 1087 publications were included in the final analytic corpus. Extracted data covered publication and citation metrics, authorship, institutional and national contributions, and keyword co-occurrence. Network and overlay visualizations were used to characterize international collaboration and temporal evolution of research themes. RESULTS: Among the 1087 included publications, annual output increased from 3 in 2010 to 198 in 2025 (partial year through November), accumulating more than 18,000 citations. The USA was the leading contributor (42%), followed by China (26%) and the United Kingdom (15%). Harvard University and the University of California, San Francisco, emerged as the most productive institutions. Methodological focus shifted from traditional ML approaches (2010-2016) to deep learning, particularly convolutional neural networks (2017-2021), and more recently to multimodal and interpretable AI (2022-2025). Research hotspots evolved from automated radiographic grading to comprehensive progression prediction integrating imaging, clinical variables, patient-reported outcomes, and pain trajectories. CONCLUSIONS: This bibliometric analysis demonstrates that AI-driven knee OA progression prediction has developed into a dynamic, globally collaborative field with growing translational focus. Emerging research hotspots suggest increasing interest in multimodal, interpretable, and patient-centered models. Key gaps include limited external validation, heavy reliance on few cohorts (OAI/MOST), and insufficient research on clinical implementation, which should be prioritized to realize AI's potential for improving patient outcomes.