Decoding high-order clinical correlations: a knowledge-driven large language model framework for specialized medical decision-making
Embedding domain-specific knowledge into LLMs may improve performance on specialized exam-style thoracic-surgery questions on this text-only benchmark, however, the present 56-item evaluation does not establish clinical equivalence, diagnostic accuracy in practice, multimodal competence, or readiness for real-world clinical decision support.