卒中影像中CT灌注软件的演变:从去卷积到人工智能
本综述回顾了CT灌注(CTP)技术原理及灌注参数估算方法,系统综述了商用CTP分析软件性能差异对神经放射学患者分层及临床结局的可能影响,并梳理了AI在CTP中的应用现状与未来挑战。目前AI在CTP估算中的应用多限于处理流程中的部分技术环节,尤以运动伪影校正为主,去卷积方法仍被广泛用于生成CTP衍生变量;AI实施的主要缺陷包括部分模型的「黑箱」性质、技术工作流实施及经济成本。
要点
- 目前AI在CTP中的应用主要限于运动伪影校正等技术环节,去卷积方法仍广泛用于生成CTP衍生变量。
- 商用CTP软件间的差异可能直接影响神经放射学患者分层及临床结局。
- 未来AI与临床数据整合有望实现患者特异性CTP图,克服基于阈值的决策局限。
核心结论
- 目前AI在CTP估算中的应用多限于处理流程中的部分技术环节,尤以运动伪影校正为主,去卷积方法仍被广泛用于生成CTP衍生变量。
- AI实施的主要缺陷包括部分模型的「黑箱」性质、技术工作流实施及经济成本。
- 未来AI与临床实践中所有可用信息的整合应致力于开发患者特异性CTP图,以克服当前基于阈值决策流程的局限,使医师更好地筛选患者并更早、更有效地治疗。
研究设计与样本
- 综述:对CTP分析软件性能进行了全面文献综述,旨在了解商用软件间可能存在的差异是否对神经放射学患者分层及临床结局有直接影响。
- 包含对CTP技术及灌注参数目前如何估算和应用于临床实践的简要技术描述。
- 提供了CTP估算软件的过去、现在和未来概览,重点为基于AI的软件。
- 讨论了技术方面及伦理考量的未来挑战。
临床应用与阅片要点
- CTP是卒中患者决策策略的主要决定因素之一,在患者分诊中非常有用。
- 商用CTP分析软件间的差异可能对神经放射学患者分层及临床结局有直接影响。
- 目前AI在CTP估算中的应用多限于处理流程中的部分技术环节,尤以运动伪影校正为主,去卷积方法仍被广泛用于生成CTP衍生变量。
- 未来AI与临床实践中所有可用信息的整合应致力于开发患者特异性CTP图,以克服当前基于阈值决策流程的局限,使医师更好地筛选患者并更早、更有效地治疗。
局限性
- AI实施的主要缺陷包括部分模型的「黑箱」性质、技术工作流实施及经济成本。
- 一些伦理、技术及厂商标准化问题仍然存在。
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
Computed tomography perfusion (CTP) represents one of the main determinants in the decision-making strategy of stroke patients, being very useful in triaging these patients. The aim of this review is to describe the current knowledge and the future applications of AI in CTP. This review contains a short technical description of the CTP technique and how perfusion parameters are currently estimated and applied in clinical practice. We then provided a comprehensive literature review on the performance of CTP analysis software aimed at understanding whether possible differences between commercially available software might have a direct implication on neuroradiological patient stratification, and therefore on their clinical outcomes. An overview of past, present, and future of software used for CTP estimation, with an emphasis on those AI-based, is provided. Finally, future challenges regarding technical aspects and ethical considerations are discussed. In the current state, most of the use of AI in CTP estimation is limited to some technical steps of the processing pipeline, and especially in the correction of motion artifacts, with deconvolution methods that are still widely used to generate CTP-derived variables. Major drawbacks in AI implementation are still present, especially regarding the "black-box" nature of some models, technical workflow implementations, and the economic costs. In the future, the integration of AI with all the information available in clinical practice should fulfill the aim of developing patient-specific CTP maps, which will overcome the current limitations of threshold-based decision-making processes and will lead physicians to better patient selection and earlier and more efficient treatments. KEY POINTS: Question AI is a widely investigated field in neuroradiology, yet no comprehensive review is yet available on its role in CT perfusion (CTP) in stroke patients. Findings AI in CTP is mainly used for motion correction; future integration with clinical data could enable personalized stroke treatment, despite ethical and economic challenges. Clinical relevance To date, AI in CTP mainly finds applications in image motion correction; although some ethical, technical, and vendor standardization issues remain, integrating AI with clinical data in stroke patients promises a possible improvement in patient outcomes.
引用
- Gragnano E, Cocozza S, Rizzuti M, et al. Evolution of CT perfusion software in stroke imaging: from deconvolution to artificial intelligence. Eur Radiol. 2025. PMID: 40628986. doi:10.1007/s00330-025-11779-8
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。