Part 1: Artificial intelligence in neurointerventional surgery: current state and challenges – a white paper

Artificial intelligence (AI) is expanding rapidly across healthcare. Given its dependence on imaging, neurointerventional surgery has seen early adoption, with convolutional neural networks deployed for large vessel occlusion detection, hemorrhage segmentation, and real-time procedural assistance. In parallel, large language models are increasingly applied to report generation, triage, and structured data extraction from unstructured clinical text. Despite this growth, widespread clinical adoption remains limited, and persistent gaps separate demonstrated promise from routine use. On the research side, the challenges are inconsistent methodological rigor and explainability, interdisciplinary fragmentation between clinicians and machine learning researchers, poor generalizability across institutions, hallucinations in generative models, algorithmic and automation bias, and clinician deskilling. On the deployment side, barriers include limited prospective validation among FDA-cleared devices, weak workflow integration, fragmented and siloed healthcare data that constrains model training and external validation, escalating cybersecurity and compliance burdens, and reimbursement frameworks built on the work relative value unit system that is not well suited for autonomous, software-driven value. Closing these gaps will require coordinated change in how the specialty develops, evaluates, and pays for clinical AI.