AURA:面向美容整形的专业人工智能临床决策支持系统
本研究开发并评估了AURA系统,它将6546篇全文开放获取的美容整形文献与商业大语言模型通过检索增强生成技术整合。在14个复杂临床场景中,系统忠实度均分0.94、答案相关性0.86、事实正确性0.77,语义评分均分0.73、最大相似度0.80。结论认为AURA是首个美容整形专用AI临床决策支持系统,能基于同行评议文献提供相关准确信息。
要点
- AURA整合6546篇全文文献与商业大语言模型,采用检索增强生成技术。
- 14个复杂场景评估:忠实度0.94、相关性0.86、事实正确性0.77。
- 成熟技术与安全考量表现优于需比较分析的新兴术式。
核心结论
- AURA是首个美容整形专用AI临床决策支持系统,能基于同行评议文献跨领域提供相关准确信息。
- 系统为美容外科医生提供高效、循证的临床决策支持资源。
研究设计与样本
- 开发并评估AURA系统,整合6546篇2001年1月至2024年9月的全文开放获取美容整形文献与商业大语言模型。
- 采用检索增强生成技术,在14个复杂临床场景中评估系统性能,涵盖面部年轻化、身体塑形、乳房手术和一般美容整形考量。
主要发现
- 忠实度均分0.94,答案相关性均分0.86,事实正确性均分0.77。
- 语义评估均分0.73(SEM评分)、0.80(SEM最大相似度),置信度评级以中等为主。
- 成熟技术和安全考量方面的表现明显优于需要比较分析的新兴术式。
关键数值与推荐
- 评估指标包括忠实度(0-1)、答案相关性(0-1)、G-Eval正确性(0-1)、语义质量评分(SEM)和SEM置信度评级。
临床怎么用
- 美容外科医生可将AURA作为高效、循证的临床决策支持资源。
- 实施时应强调透明呈现来源,以补充外科专业知识。
局限性
- 未来开发应聚焦扩展知识来源和开展前瞻性临床验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Aesthetic surgery requires integration of specialized knowledge with clinical expertise, yet traditional literature search methods are time-intensive when addressing specific clinical questions. Although artificial intelligence (AI) has transformed various aspects of healthcare delivery, no AI clinical decision support systems have been specifically developed to enhance evidence-based practice in aesthetic surgery. OBJECTIVES: The authors aim to develop and evaluate Aesthetic Surgery Using Retrieval Augmentation (AURA), a novel AI-powered clinical decision support system designed specifically for evidence-based guidance in aesthetic surgery. METHODS: AURA integrates a comprehensive database of 6546 full-text open-access aesthetic surgery publications (January 2001-September 2024) with a commercial large language model using retrieval-augmented generation technology. System performance was rigorously assessed across 14 complex clinical scenarios spanning facial rejuvenation, body contouring, breast procedures, and general aesthetic surgery considerations. Evaluation metrics included faithfulness to source materials (0-1), answer relevancy (0-1), G-Eval correctness (0-1), semantic quality scores (SEM), and SEM confidence ratings. RESULTS: AURA demonstrated exceptional performance with mean scores of 0.94 for faithfulness, 0.86 for answer relevancy, and 0.77 for factual correctness. Semantic evaluation revealed strong results with average scores of 0.73 (SEM score) and 0.80 (SEM max similarity), predominantly with moderate confidence ratings. Performance was notably stronger for established techniques and safety considerations compared with emerging procedures requiring comparative analysis. CONCLUSIONS: In this study, the authors present AURA, the first specialized AI clinical decision support system for aesthetic surgery. This novel system effectively delivers relevant, accurate information across diverse aesthetic surgery domains based on peer-reviewed literature. AURA offers aesthetic surgeons an efficient, evidence-based resource for clinical decision support. Future development should focus on expanding knowledge sources and prospective clinical validation, with implementation emphasizing transparent source presentation to complement surgical expertise.
引用
- Ozmen BB, Singh N, Shah K, et al. Specialized Artificial Intelligence Clinical Decision Support System for Aesthetic Surgery: Aesthetic Surgery Using Retrieval Augmentation (AURA). Aesthet Surg J. 2025. PMID: 40579186. doi:10.1093/asj/sjaf120
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。