科研动态PubMed 摘要

人工智能在颅内动脉瘤诊断与结局预测中的应用:一项荟萃分析与综述。

PubMed 学科轮询 · 2026-07-31

完整摘要 · 中文译文

背景:颅内动脉瘤(IAs)影响高达5%的普通人群,若未被发现或分类错误,则构成蛛网膜下腔出血的显著风险。人工智能(AI)的最新进展,尤其是深度学习,已在增强动脉瘤检测和结局预测方面显示出前景。我们开展了一项系统综述和荟萃分析,以评估AI在IA检测中的诊断性能,并综述其在破裂风险和结局预测中的应用。 方法:遵循PRISMA指南,我们筛选了主要数据库中报告基于AI的IA检测和/或结局预测的研究。符合条件的研究包括AI与临床医生或AI增强判读的比较。计算了合并敏感性、特异性和评估者间一致性(Fleiss' κ)。结局预测研究采用叙述性综述。 结果:二十项研究(n = 20)符合荟萃分析的纳入标准。AI的合并敏感性和特异性分别为0.96和0.95。然而,观察到研究间存在显著异质性。因此,这些估计值应被解释为对所报告研究性能的探索性总结,而非真实世界诊断准确性的确定性指标。亚组分析显示,AI对> 3 mm动脉瘤具有稳健的性能(敏感性:0.98),而对< 3 mm病变的性能下降(敏感性:0.75)。AI辅助将临床医生敏感性提高至0.92,并将评估者间一致性从κ = 0.67提高至0.86。对结局预测模型的综述显示有前景的AUC(高达0.96)。 用于预测破裂风险、动脉瘤生长及血管痉挛等并发症。 结论:人工智能在颅内动脉瘤检测和结局预测方面表现出良好前景,尤其是当其作为决策支持工具与临床医生配合使用时。尽管仍存在局限性——特别是在检测微小动脉瘤和跨中心泛化性方面——人工智能仍是神经血管诊断中有价值的辅助手段。未来方向包括改进训练数据集、多中心验证以及无缝临床整合,以充分发挥人工智能辅助神经影像和风险分层的潜力。 临床试验编号:不适用。

完整摘要 · 英文原文

BACKGROUND: Intracranial aneurysms (IAs) affect up to 5% of the general population and pose a significant risk of subarachnoid hemorrhage if undetected or misclassified. Recent advances in artificial intelligence (AI), particularly deep learning, have shown promise in enhancing aneurysm detection and outcome prediction. We conducted a systematic review and meta-analysis to evaluate the diagnostic performance of AI in IA detection and reviewed its applications in rupture risk and outcome prediction. METHODS: Following PRISMA guidelines, we screened major databases for studies reporting AI-based IA detection and/or outcome prediction. Eligible studies included comparisons of AI versus clinicians or AI-augmented interpretations. Pooled sensitivity, specificity, and interrater agreement (Fleiss' κ) were calculated. Outcome prediction studies were reviewed narratively. RESULTS: Twenty studies (n = 20) met the inclusion criteria for meta-analysis. Pooled AI sensitivity and specificity were 0.96 and 0.95 respectively. However, substantial heterogeneity across studies was observed. These estimates should therefore be interpreted as exploratory summaries of reported study performance rather than definitive measures of real-world diagnostic accuracy. Subgroup analysis revealed robust AI performance for aneurysms > 3 mm (sensitivity: 0.98), while performance declined for < 3 mm lesions (sensitivity: 0.75). AI assistance improved clinician sensitivity to 0.92 and increased interrater agreement from κ = 0.67 to 0.86. Review of outcome prediction models showed promising AUCs (up to 0.96) for predicting rupture risk, aneurysm growth, and complications like vasospasm. CONCLUSIONS: AI shows promising performance in intracranial aneurysm detection and outcome prediction, particularly when used as a decision-support tool alongside clinicians. While limitations remain-especially in detecting very small aneurysms and generalizability across centers-AI represents a valuable adjunct in neurovascular diagnostics. Future directions include improved training datasets, multicenter validation, and seamless clinical integration to realize the full potential of AI-assisted neuroimaging and risk stratification. CLINICAL TRIAL NUMBER: Not applicable.

English
Artificial intelligence in the diagnosis and outcome prediction of intracranial aneurysms: a meta-analysis and review.
内容仅供参考,不构成诊疗建议,不能替代执业医师的诊断。

← 返回元界