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

Abstract

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.

Ähnliche Arbeiten