创伤外科中的生成式人工智能与语言模型：在临床、科研和教学中的应用

BACKGROUND: Generative artificial intelligence (AI) and large language models (LLM) are evolving from general purpose text assistants toward increasingly contextualized, multimodal systems integrated into clinical workflows.

OBJECTIVE: Presentation of current applications of generative AI and LLMs in trauma surgery, with a focus on clinical care and additional aspects of research and teaching.

MATERIAL AND METHODS: Narrative literature review based on a MEDLINE/PubMed search and supplementary hand search, including primary regulatory sources.

RESULTS: The most immediate clinical potential currently lies in documentation, information processing and patient communication. Initial studies demonstrate relevant time savings with AI-assisted documentation. Current systems achieve high performance in some clearly defined triage and decision-support tasks, whereas complex individualized and multimodal applications show lower and more heterogeneous reliability. In research and teaching LLMs can support standardized workflows, case generation and learning activities; however, the available evidence is still frequently based on retrospective or simulated studies.

CONCLUSION: The safe implementation requires validation of the specific AI system in its intended context of use, reliable knowledge sources, sufficient clinical context, data protection and effective human control. As the clinical consequences of AI-generated outputs increase, so do the requirements for validation and physician responsibility in making decisions.