Comparative accuracy of large language models and orthopaedic clinicians in knee osteoarthritis and tibial plateau fracture classification systems.

BACKGROUND: Multimodal large language models (LLMs) can process radiographic images, but their accuracy across orthopaedic classification systems requires further evaluation. This study compared the classification accuracy of four LLMs with that of orthopaedic surgeons and residents. METHODS: This cross-sectional study included 100 cases obtained from Radiopaedia, with 20 cases each for the Kellgren-Lawrence, Ahlback, Schatzker, Hohl and Moore, and AO Foundation/Orthopaedic Trauma Association (AO/OTA) classifications. Each case comprised anteroposterior and lateral radiographs. Five orthopaedic surgeons, five residents, and four LLMs assessed the same cases, yielding 1400 classification assessments. Three independent senior orthopaedic surgeons established the consensus reference standard. AI evaluations were conducted in April 2026 using a standardized Turkish-language prompt, without classification definitions or examples. Only the first response was recorded. Analyses were descriptive because paired case-level classifications were unavailable. RESULTS: Overall accuracy was 91.2% (456/500) for orthopaedic surgeons, 70.2% (351/500) for residents, and 51.0% (204/400) for LLMs. LLM accuracy was 70.0% for Kellgren-Lawrence, 68.8% for Ahlback, 67.5% for AO/OTA, 28.8% for Hohl and Moore, and 20.0% for Schatzker. Model-specific accuracy was 58.0% for Claude Opus 4.1, 57.0% for GPT-5.5, 45.0% for Gemini 2.5 Pro, and 44.0% for DeepSeek-VL2. Orthopaedic surgeons achieved the highest observed accuracy across all five classification systems. CONCLUSION: LLM classification accuracy varied across systems and was lower overall than that of orthopaedic surgeons and residents. These findings are limited to the selected educational dataset and evaluation conditions. They do not establish educational or clinical utility; validation in larger, representative datasets is required.