脑小血管病（CSVD）是缺血性脑卒中与血管性认知障碍的主要病因，其发生发展与颅内大动脉到远端微循环的多尺度血流动力学功能障碍有密切关系。然而，传统的影像学评估方法还难以做到对这一完整通路的精准、无创、一体化定量分析。近年来，人工智能（AI）技术的快速发展为解决这一难题提供了革新性工具。本文系统综述了AI赋能CSVD血流动力学评估的研究进展与临床转化现状，阐述了AI在多模态影像数据自动化分割、降噪、配准等预处理环节中的技术支撑作用；分析AI在颅内大动脉血流速度与壁面剪切应力、脑微循环灌注异质性、脑自动调节功能、穿支动脉三维形态及血脑屏障渗透性等关键参数精准量化中的应用进展；总结AI通过多尺度信息融合在CSVD亚型鉴别、认知下降与疾病进展风险分层中的预测价值。在此基础上，进一步探讨当前临床转化的核心瓶颈，包括可解释性AI技术缺乏结构化的人因可用性验证、后验解释方法可能诱发自动化偏差、传统循证医学框架与动态演进型AI系统的范式错配，以及卫生技术评估证据链的断层。未来突破需聚焦构建“替代终点-中间指标-硬终点”分层验证路径，将人因验证前置于模型开发流程，以及建立覆盖全生命周期的动态监管评估体系。

Cerebral small vessel disease (CSVD) is the primary cause of ischemic stroke and vascular cognitive impairment, and its occurrence and development are closely related to multi-scale hemodynamic dysfunction from intracranial large arteries to distal microcirculation. However, it is difficult for traditional imaging assessment methods to achieve precise, non-invasive, and integrated quantitative analysis of this complete pathway. In recent years, the rapid development of artificial intelligence (AI) technology has provided innovative tools to solve this problem. This article systematically reviews the research progress and clinical translation status of AI-empowered hemodynamic assessment of CSVD. First, we elaborate on the technical support role of AI in the pre-processing steps such as automated segmentation, noise reduction, and registration of multimodal imaging data. Then, we focus on the application progress of AI in the precise quantification of key parameters such as intracranial large artery blood flow velocity and wall shear stress, cerebral microcirculation perfusion heterogeneity, cerebral autoregulatory function, three-dimensional morphology of perforating arteries, and blood-brain barrier permeability. Finally, we summarize the predictive value of AI through multi-scale information fusion in the subtype differentiation of CSVD, cognitive decline, and risk stratification of disease progression. Based on this, we further explore the current core bottlenecks in clinical translation, including: the lack of structured human-factor usability verification for explainable AI technology, the potential for automated biases triggered by post hoc explanations, the mismatch between the traditional evidence-based medical framework and the dynamic evolving AI system paradigm, and the discontinuity of the evidence chain of health technology assessment. Future breakthroughs need to focus on constructing a "surrogate endpoint - intermediate indicator - hard endpoint" hierarchical verification path, placing human-factor verification prior to the AI model development process, and establishing a dynamic regulatory assessment system covering the entire life cycle.