Artificial Intelligence in Cerebrovascular Disease Management: A Comprehensive Review of Risk Prediction, Diagnosis, Therapeutic Optimization, and Clinical Translation.

Cerebrovascular diseases (CVDs) impose a heavy global health burden, necessitating efficient management strategies. Artificial intelligence (AI) has become a key transformative tool across the CVD care continuum, and this review systematically synthesizes AI's latest advancements, limitations, and clinical translation pathways in CVD management, adhering to PRISMA-ScR guidelines. A literature search was performed in four core databases (PubMed, Web of Science, EMBASE, IEEE Xplore) for studies published between 2018-2023. After strict screening (inclusion: original research/clinical trials with clear indicators; exclusion: unvalidated studies/conference abstracts), 128 high-quality studies were included, with quality assessed via NOS and QUADAS-2. Key AI applications in CVD management include: (1) Risk prediction: Multimodal models (radiomics-CFD, EHR-imaging) achieve AUC >0.9, but performance declines in elderly patients (>75 years, ΔAUC=0.08-0.12); (2) Diagnosis: Systems like Viz LVO and DeepHemorrhage reduce LVO detection time to 6 minutes and hemorrhage segmentation Dice to 0.94, yet face false positives (3.5-5%) and workflow delays; (3) Therapeutic optimization: Intraoperative AI (eg, Siemens AI-Path) shortens microcatheter placement time by 61%, and pharmacogenomic models cut antiplatelet complications by 37%; (4) Long-term monitoring: Mobile platforms (eg, NeuroVision™) automate NIHSS scoring (ICC=0.93) but lose accuracy in home settings (ICC=0.85-0.88). Critical limitations of current AI include single-center data bias, poor interpretability, and legal risks (unclear misdiagnosis liability). This review proposes three innovative solutions: a "data-model-clinical" closed loop, a multidimensional AI value evaluation system, and defining the "human-AI collaboration boundary" in neurointerventions. Future directions focus on primary care-adapted lightweight models, comorbidity-specific algorithms, and AI-assisted rehabilitation. This review emphasizes that physician-AI collaboration and standardized frameworks (eg, AI-RADS, WHO-ITU guidelines) are critical for AI's sustainable translation in CVD care. Addressing current gaps will enable AI to further improve therapeutic efficiency and functional outcomes, alleviating the global CVD burden.