代理式人工智能在放射学中的应用综述:从当前临床整合到未来创新
归纳出5个主题:自主临床决策支持、工作流编排、多模态图像分析、报告与沟通、伦理指导。多数纳入研究为技术性或探索性,亟需前瞻性真实世界应用研究评估临床工作流整合。
要点
- 代理式AI可自主执行处方任务,有望提升工作流效率和报告流程。
- 当前证据以技术开发和早期探索为主,临床整合证据有限。
- 需前瞻性真实世界研究验证代理式AI在放射科工作流中的效果。
英文原文(摘要)· 点击展开
AIM: Artificial intelligence (AI) is one of the most revolutionary developments in the field of medicine in recent history, with radiology being one of the strongest beneficiaries. AI models predominantly relied on user input to generate 'human-like' responses through a series of algorithms. Newer developments in this domain include agentic AI systems, where individual AI systems work on prescribed tasks. This study reviews the current evidence base to provide a synthesis of the present literature.
METHODS: For this scoping review, parallel searches of PubMed, Embase, Web of Science, and Scopus were conducted for all papers on the theme of agentic AI in radiology published between January 2015 and October 2025. Papers were then screened by two independent reviewers, with conflicts resolved through consensus. Data were extracted from papers according to a predetermined data extraction table and was grouped by common themes to provide a synthesis of the current evidence base.
RESULTS: Searches yielded a total of 129 articles, 27 of which were included in the final review after screening. There were 5 main themes identified across the 27 studies: the role of agentic AI in autonomous clinical decision support; workflow orchestration; multimodal systems for image analysis; reporting and communications; and ethical guidance. Across all included studies, many were technical papers or exploratory, highlighting the need for prospective real-world application studies to assess integration into clinical workflows.
CONCLUSION: Agentic AI provides an exciting and novel way to improve workflow efficiency and streamline reporting pipelines.
引用
- Chinniah P, Gibson J, Chapala S, et al. Review of agentic artificial intelligence (AI) in radiology: from current clinical integration to future innovations. Clin Radiol. 2026. PMID: 42048905. doi:10.1016/j.crad.2026.107351
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。