人工智能在耳鼻咽喉头颈外科中的应用：当前应用、局限性与未来展望。

文献信息

中文摘要

目的： 人工智能（AI）正日益融入现代耳鼻咽喉头颈外科实践，并已成为当代医学中发展最迅速的技术之一。机器学习、深度学习、计算机视觉和多模态AI系统的最新进展，加速了耳鼻咽喉头颈外科多个亚专科诊断和治疗应用的发展。
方法： 本叙述性综述通过检索PubMed/MEDLINE、Scopus和Web of Science数据库的结构化文献检索完成。使用预定义关键词的组合识别评估人工智能在耳鼻咽喉头颈外科中应用的相关出版物。优先考虑以英文发表的同行评议研究，这些研究探讨了人工智能在耳鼻咽喉头颈外科中具有临床相关性的诊断、预后、手术、教育或工作流程相关应用。回顾性研究和前瞻性研究、综述文章、具有里程碑意义的方法学研究以及对该领域具有重大影响的代表性出版物均被纳入考虑。
结果： AI辅助技术在诊断成像、内镜评估、听力学、鼻科学、喉科学、前庭医学、手术模拟、头颈肿瘤学、放疗计划和预测分析中显示出有前景的临床效用。新出现的证据表明，深度学习算法可能增强鼻窦疾病的检测，促进自动图像分割，预测淋巴结转移，分类甲状腺结节，并支持头颈癌患者的预后建模。此外，整合放射学、病理学、分子和临床数据的多模态AI系统可能进一步提高诊断精确度和个性化治疗计划。生成式AI工具，包括基于GPT的大语言模型，也已在医学教育、图像解读、鉴别诊断和临床决策支持中显示出新兴应用。
结论： 然而，重要的方法学和转化挑战继续限制更广泛的临床实施。当前挑战包括外部验证有限、回顾性研究设计、数据集异质性、算法偏倚、缺乏透明度、患者隐私问题、医疗法律不确定性以及与自动化偏倚相关的潜在风险。本叙述性综述总结了人工智能在耳鼻咽喉头颈外科中的当前临床应用，讨论了现有技术的优势和局限性，并强调了负责任临床整合的未来展望。当前证据表明，AI可能作为耳鼻咽喉头颈外科实践中有价值的辅助工具；然而，在实现广泛采用之前，仍需进一步的前瞻性多中心研究、标准化验证框架和伦理监督。

英文摘要

PURPOSE: Artificial intelligence (AI) is increasingly integrated into modern otolaryngology practice and has emerged as one of the most rapidly evolving technologies in contemporary medicine. Recent advances in machine learning, deep learning, computer vision, and multimodal AI systems have accelerated the development of diagnostic and therapeutic applications across multiple otolaryngology subspecialties.
METHODS: This narrative review was conducted through a structured literature search of PubMed/MEDLINE, Scopus, and Web of Science databases. Relevant publications evaluating artificial intelligence applications in otolaryngology were identified using combinations of predefined keywords. Priority was given to peer-reviewed studies published in English that investigated clinically relevant diagnostic, prognostic, surgical, educational, or workflow-related applications of artificial intelligence within otolaryngology. Both retrospective and prospective studies, review articles, landmark methodological studies, and representative publications with substantial influence on the field were considered.
RESULTS: AI-assisted technologies have demonstrated promising clinical utility in diagnostic imaging, endoscopic assessment, audiology, rhinology, laryngology, vestibular medicine, surgical simulation, head and neck oncology, radiotherapy planning, and predictive analytics. Emerging evidence suggests that deep learning algorithms may enhance detection of sinonasal disease, facilitate automated image segmentation, predict lymph node metastasis, classify thyroid nodules, and support prognostic modeling in head and neck cancer patients. In addition, multimodal AI systems integrating radiologic, pathologic, molecular, and clinical data may further improve diagnostic precision and personalized treatment planning. Generative AI tools, including GPT-based large language models, have also demonstrated emerging applications in medical education, image interpretation, differential diagnosis, and clinical decision support.
CONCLUSION: Nevertheless, important methodological and translational challenges continue to limit broader clinical implementation. Current challenges include limited external validation, retrospective study design, dataset heterogeneity, algorithmic bias, lack of transparency, patient privacy concerns, medico-legal uncertainty, and the potential risks associated with automation bias. This narrative review summarizes current clinical applications of artificial intelligence in otolaryngology, discusses the strengths and limitations of currently available technologies, and highlights future perspectives for responsible clinical integration. Current evidence suggests that AI may serve as a valuable adjunctive tool in otolaryngology practice; however, further prospective multicenter studies, standardized validation frameworks, and ethical oversight remain necessary before widespread adoption can be achieved.