文献速递临床血液与体液质量管理与危急值分子诊断

人工智能与机器学习在血栓与止血领域的范围综述

Artificial intelligence and machine learning in thrombosis and hemostasis: a scoping review of clinical and laboratory applications, challenges, and future directions.
PubMed 文献 · Clin Chem Lab Med · 2025-12-17

本范围综述遵循PRISMA-ScR指南,检索MEDLINE、EMBASE、Web of Science和Scopus(2020-2025年),纳入107项原始研究,系统梳理AI与ML在血栓与止血临床和实验室诊断中的应用、挑战与未来方向。

要点

  • 临床应用集中于VTE、PE、DVT预测建模、抗凝管理和疾病风险分层。
  • 实验室应用较少,涉及自动化质控、凝块检测和检测解读增强。
  • 主要局限包括回顾性单中心小样本、缺乏外部验证和模型可解释性差。

核心结论

  • AI与ML在血栓与止血领域有改善诊断准确性和临床决策的潜力,但广泛采用仍面临数据、验证、可解释性和跨学科协作等障碍。

研究设计与样本

  • 遵循PRISMA-ScR指南的范围综述,检索MEDLINE、EMBASE、Web of Science和Scopus(2020-2025年),纳入107项原始研究进行分析。

主要发现

  • 临床应用主要集中于静脉血栓栓塞症、肺栓塞、深静脉血栓的预测建模、抗凝管理和疾病风险分层,采用神经网络、随机森林和梯度提升等算法。
  • 实验室AI应用较少,但提供了自动化质量控制、凝块检测和检测解读增强,可能改善决策。
  • 纳入研究的主要局限包括依赖回顾性、单中心、小样本数据集,有限的外部验证,模型可解释性担忧以及临床工作流整合挑战。
  • 血栓与止血领域专家与AI-ML专家之间持续存在的跨学科脱节,加上监管障碍、数据碎片化和劳动密集型数据标注过程,是更广泛采用的主要障碍。

临床怎么用

  • 临床可关注AI在VTE、PE、DVT预测和抗凝管理中的潜在应用,但需注意现有证据多来自回顾性单中心小样本研究。
  • 实验室可探索AI用于自动化质控、凝块检测和检测解读,但应重视外部验证和模型可解释性。
  • 推动实验室科学家与AI-ML专家协作,建立结构化教育项目和监管框架,以促进AI在血栓与止血领域的安全有效应用。

局限性

  • 纳入研究多依赖回顾性、单中心、小样本数据集,外部验证有限。
  • 模型可解释性 concerns 和临床工作流整合挑战突出。
  • 跨学科脱节、监管障碍、数据碎片化和劳动密集型数据标注阻碍广泛采用。

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

This scoping review followed the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines to systematically map the current landscape of artificial intelligence (AI) and machine learning (ML) applications in the field of thrombosis and hemostasis (T&H), specifically targeting diagnostic enhancements in clinical and laboratory settings. Utilizing comprehensive searches across MEDLINE, EMBASE, Web of Science, and Scopus (2020-2025), 107 original studies met inclusion criteria and were analyzed. Clinical applications predominantly focused on predictive modelling for venous thromboembolism (VTE), pulmonary embolism (PE), deep vein thrombosis (DVT), anticoagulant management, and disease risk stratification, employing algorithms including neural networks, random forests, and gradient boosting. Laboratory-based AI implementations, though fewer, provided automated quality control, clot detection, and assay interpretation enhancements for potential better decision-making. Significant limitations addressed by the include studies include reliance on retrospective, single-center, small-sample datasets, limited external validation, model interpretability concerns, and integration challenges into clinical workflows. Persistent interdisciplinary disconnect between hemostasis domain experts and AI-ML specialists, compounded by regulatory hurdles, fragmented data, and labor-intensive data labelling processes, was highlighted as a major barrier to broader adoption. Recommendations for future research include developing large, externally validated multicenter datasets, transparent and interpretable ML models, prospective clinical validations, and user-centered integration strategies. Enhancing collaboration between laboratory scientists and AI-ML experts, establishing structured education programs, and creating regulatory frameworks are essential next steps to fully realize the potential of AI for significantly improving diagnostic accuracy, clinical decision-making, and patient management in T&H.

来源:Clin Chem Lab Med,查看原文。版权归原作者所有。

引用

  1. Altememi MA, Favaloro EJ, Islam MZ, et al. Artificial intelligence and machine learning in thrombosis and hemostasis: a scoping review of clinical and laboratory applications, challenges, and future directions. Clin Chem Lab Med. 2025. PMID: 41400592. doi:10.1515/cclm-2025-1450

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。