人工智能在慢性鼻窦炎中的应用：从分子分型到个性化临床管理——一项系统综述。

文献信息

中文摘要

背景： 慢性鼻窦炎（CRS）是一种异质性炎症性疾病，造成沉重的健康和经济负担。人工智能（AI）已成为应对临床和研究环境中CRS复杂性的一种有前景的工具。方法：我们按照PRISMA指南进行了一项系统综述，检索PubMed、Web of Science和Embase中利用机器学习、深度学习和自然语言处理的同行评审文章，重点关注CRS的分子水平或免疫学机制。结果：共纳入31项研究进行定性综合。我们的结果表明，尽管AI驱动的分析有助于对患者内型进行分层，但所识别的分型方案在聚类数量和特征上存在显著异质性，引发对其可重复性的担忧。AI应用还包括组织病理学图像解读、药物发现、生物制剂治疗选择以及评估翻修手术风险和嗅觉功能障碍的预后建模。然而，使用PROBAST+AI框架，我们发现87.1%的受评估研究存在高总体偏倚风险，主要由于依赖表观性能或内部验证以及对模型性能的独立评估有限，表明许多报告的性能估计可能过于乐观。结论：未来需要大规模、多中心合作，将AI整合到临床工作流程中，以建立可靠且可推广的AI应用，用于CRS的个性化管理。

英文摘要

Chronic rhinosinusitis (CRS) is a heterogeneous inflammatory disease imposing a substantial health and economic burden. Artificial intelligence (AI) has emerged as a promising tool to address CRS complexity in both clinical and research settings. We conducted a systematic review following PRISMA guidelines, searching PubMed, Web of Science, and Embase for peer-reviewed articles utilizing machine learning, deep learning, and natural language processing, with a focus on molecular-level or immunologic mechanisms in CRS. Thirty-one studies were included in the qualitative synthesis. Our results demonstrate that while AI-driven analysis helps stratify patient endotypes, the identified schemes exhibit substantial heterogeneity in cluster numbers and features, raising concerns about their reproducibility. AI applications also included histopathological image interpretation, drug discovery, biologic therapy selection, and prognostic modeling to evaluate revision surgery risks and olfactory dysfunction. However, using the PROBAST+AI framework, we found that 87.1% of the assessed studies carry a high overall risk of bias, mainly due to the reliance on apparent performance or internal validation and limited independent evaluation of model performance, indicating that many reported performance estimates may be optimistic. Future large-scale, multicenter collaborations integrating AI into clinical workflows will be essential to establish reliable and generalizable AI applications for personalized CRS management.