分类缺血性卒中发病时间的最佳诊断方法:一项系统综述与Meta分析
该系统综述与Meta分析纳入截至2025年7月的9030例卒中患者,比较不同诊断方法对缺血性卒中发病时间的分类性能。结果显示,DWI-FLAIR不匹配的人工判读(现行金标准)AUC为0.71、敏感度0.62、特异度0.78,分类性能中等;基于CT影像组学特征的机器学习模型AUC为0.89、敏感度0.85、特异度0.86,在综述所纳入的模型中分类性能最佳。
要点
- DWI-FLAIR不匹配人工判读识别发病时间性能中等(AUC 0.71)。
- 基于CT影像组学特征的机器学习模型分类性能最佳(AUC 0.89)。
核心结论
- 机器学习模型在分类缺血性卒中发病时间方面优于DWI-FLAIR不匹配的人工判读;基于CT影像组学数据的高效AI模型可能获得最佳分类性能,以判定患者是否符合静脉tPA治疗条件并改善治疗结局。
研究设计与样本
- 系统综述与Meta分析,检索Web of Science、PubMed、Scopus、Embase和Cochrane数据库至2025年7月,共纳入9030例卒中患者。
诊断标准与关键数值
- DWI-FLAIR不匹配人工判读(现行金标准方法)识别发病时间:AUC=0.71(95%CI: 0.66-0.76),敏感度=0.62(95%CI: 0.54-0.71),特异度=0.78(95%CI: 0.72-0.84)。
- 基于CT数据影像组学特征的机器学习模型分类发病时间:AUC=0.89(95%CI: 0.80-0.98),敏感度=0.85(95%CI: 0.75-0.96),特异度=0.86(95%CI: 0.73-1.00)。
临床应用与阅片要点
- 机器学习模型在分类发病时间上优于DWI-FLAIR不匹配人工判读。
- 基于CT影像组学数据的高效AI模型可能获得最佳分类性能,以判定患者是否符合静脉tPA治疗条件并改善治疗结局。
局限性
- 原文未报告局限性。
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)。
英文原文(摘要)· 点击展开
BACKGROUND: The success of intravenous thrombolysis with tPA (IV-tPA) as the fastest and easiest treatment for stroke patients is closely related to time since stroke onset (TSS). Administering IV-tPA after the recommended time interval (< 4.5 h) increases the risk of cerebral hemorrhage. Despite advances in diagnostic approaches have been made, the determination of TSS remains a clinical challenge. In this study, the performances of different diagnostic approaches were investigated to classify TSS.
MATERIALS AND METHODS: A systematic literature search was conducted in Web of Science, Pubmed, Scopus, Embase, and Cochrane databases until July 2025. The overall AUC, sensitivity, and specificity magnitudes with their 95%CIs were determined for each diagnostic approach to evaluate their classification performances.
RESULTS: This systematic review retrieved a total number of 9030 stroke patients until July 2025. The results showed that the human readings of DWI-FLAIR mismatch as the current gold standard method with AUC = 0.71 (95%CI: 0.66-0.76), sensitivity = 0.62 (95%CI: 0.54-0.71), and specificity = 0.78 (95%CI: 0.72-0.84) has a moderate performance to identify the TSS. ML model fed by radiomic features of CT data with AUC = 0.89 (95%CI: 0.80-0.98), sensitivity = 0.85 (95%CI: 0.75-0.96), and specificity = 0.86 (95%CI: 0.73-1.00) has the best performance in classifying TSS among the models reviewed.
CONCLUSION: ML models fed by radiomic features better classify TSS than the human reading of DWI-FLAIR mismatch. An efficient AI model fed by CT radiomic data could yield the best classification performance to determine patients' eligibility for IV-tPA treatment and improve treatment outcomes.
引用
- Zakariaee SS, Kadir DH, Molazadeh M, et al. The best diagnostic approach for classifying ischemic stroke onset time: A systematic review and meta-analysis. Neuroradiology. 2025. PMID: 40938372. doi:10.1007/s00234-025-03745-4
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。