Part 2: Toward responsible AI adoption in neurointerventional surgery

Closing the gap between artificial intelligence’s (AI’s) demonstrated promise in neurointerventional surgery and its routine clinical use requires coordinated change across research, deployment and reimbursement. On the research side, progress requires interdisciplinary collaboration between clinicians and machine learning specialists, rigorous exploratory data analysis prior to modelling, internal, temporal, and external validation, and adherence to AI-specific reporting standards. On the deployment side, physician augmentation with AI should be done through defined oversight modes (second-reader, triage, decision support and intraprocedural assistance), mitigation of generative-model failure modes such as hallucinations, and training that preserves the competencies clinicians need to recognise when AI is wrong. On the reimbursement side, sustainable adoption requires anchoring in demonstrated clinical and economic value through Current Procedural Terminology codes or New Technology Add-on Payment and Transitional Coverage for Emerging Technologies pathways rather than retrofits of the wRVU system. Underlying all is the need for centralised, multi-institutional data repositories that treat clinical data as a public good rather than a monetisable institutional asset.