Scholarlink

<p>Large Language Models (LLMs) unlock new possibilities for the automated processing and generation of natural language in clinical medicine. The aim of this narrative review was to summarize the current literature on the applications of LLMs in vascular surgery and to present their benefits, limitations, and challenges from a&#xa0;clinical perspective. A&#xa0;PubMed search identified nine original studies, which were categorized into three areas of application: (1)&#xa0;decision support and diagnostics, (2)&#xa0;documentation and data extraction, and (3)&#xa0;patient information and communication. In simulation-based scenarios, LLMs were able to reliably identify vascular emergencies and prioritize differential diagnoses. Retrospective studies demonstrated high accuracy in extracting standardized data from narrative surgical reports and radiological findings. In patient communication, frequently asked questions were answered mostly correctly and comprehensibly, although linguistic complexity and occasional factual errors were reported as limitations. Key challenges relate to the lack of external validation of the models, the risk of factually incorrect but plausible-sounding answers (so-called “hallucinations”), the technical integration into existing IT infrastructures, and data-protection requirements related to the processing of personal data. The latter necessitate local, GDPR-compliant infrastructures and careful consideration of privacy-preserving artificial intelligence. In addition to the positive-use cases, the need for robust control mechanisms and targeted implementation strategies is also emphasized.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Einsatz von Large Language Models (LLMs) in der Gefäßchirurgie: Möglichkeiten und Limitationen

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月1日

查看全文
  • Johannes Hatzl,
  • Oana Bartos,
  • Christian-Alexander Behrendt,
  • Bernhard Dorweiler,
  • Jörg Heckenkamp,
  • Grischa Hoffmann,
  • Benedikt Reutersberg,
  • Christian Uhl

摘要

Large Language Models (LLMs) unlock new possibilities for the automated processing and generation of natural language in clinical medicine. The aim of this narrative review was to summarize the current literature on the applications of LLMs in vascular surgery and to present their benefits, limitations, and challenges from a clinical perspective. A PubMed search identified nine original studies, which were categorized into three areas of application: (1) decision support and diagnostics, (2) documentation and data extraction, and (3) patient information and communication. In simulation-based scenarios, LLMs were able to reliably identify vascular emergencies and prioritize differential diagnoses. Retrospective studies demonstrated high accuracy in extracting standardized data from narrative surgical reports and radiological findings. In patient communication, frequently asked questions were answered mostly correctly and comprehensibly, although linguistic complexity and occasional factual errors were reported as limitations. Key challenges relate to the lack of external validation of the models, the risk of factually incorrect but plausible-sounding answers (so-called “hallucinations”), the technical integration into existing IT infrastructures, and data-protection requirements related to the processing of personal data. The latter necessitate local, GDPR-compliant infrastructures and careful consideration of privacy-preserving artificial intelligence. In addition to the positive-use cases, the need for robust control mechanisms and targeted implementation strategies is also emphasized.