Publicly available multimodal large language models for ocular surface infections: benchmarking against corneal specialists in triage, diagnosis and treatment

Key Points

• To assess the effectiveness of multimodal large language models in diagnosing ocular surface infections compared to corneal specialists.
• Benchmarking multimodal LLMs against expert evaluations.
• Incorporating clinical context and slit-lamp images for analysis.
• Evaluating therapeutic reasoning and pathogen identification.
• Multimodal LLMs achieved near-expert-level diagnosis and triage performance.
• Identified gaps in therapeutic reasoning and rare pathogen recognition.
• Potential for these models to support care in resource-limited settings.

Abstract