指南与共识儿科影像技术与安全

人工智能在儿科放射学中实施以保障患者安全:来自ACR、ESPR、SPR、SLARP、AOSPR、SPIN的多学会声明

Artificial Intelligence Implementation in Pediatric Radiology for Patient Safety: A Multisociety Statement From the ACR, ESPR, SPR, SLARP, AOSPR, SPIN.
指南 / 共识 · J Am Coll Radiol · 2025-11-25

该多学会立场声明指出当前人工智能解决方案与指南主要针对成人,忽视儿童在生理、发育阶段和临床需求上的差异;系统提出监管与采购、实施与整合、解读与上市后监测、教育四大支柱,并倡导以儿童为中心的AI整合框架,强调儿科特异性安全评级、数据集多样性、可量化透明度指标与模型可解释性,推荐分阶段实施策略与持续上市后监测。

要点

  • 声明提出儿科特异性安全评级、纳入多样化儿科人群数据集、可量化透明度指标和模型可解释性,以减轻偏倚并确保AI系统适用于儿童。
  • 监管与采购的重要步骤包括风险评估、数据集多样性、透明度和网络安全。
  • 推荐分阶段实施策略,包括早期试点测试、利益相关者参与以及持续监测既定性能基准的全面上市后监测。

核心结论

  • 倡导以儿童为中心的AI整合框架,优先考虑儿童的独特需求并保障其安全、准确性和整体福祉。

研究设计与样本

  • 由ACR、ESPR、SPR、SLARP、AOSPR、SPIN共同发布的多学会立场声明,系统阐述AI采用的四大支柱:监管与采购、实施与整合、解读与上市后监测、教育。

诊断标准与关键数值

  • 提出儿科特异性安全评级、纳入多样化儿科人群数据集、可量化透明度指标和模型可解释性,以减轻偏倚并确保AI系统适用于儿童。

临床应用与阅片要点

  • 监管与采购阶段应重视风险评估、数据集多样性、透明度和网络安全。
  • 实施阶段推荐分阶段策略,包括早期试点测试、利益相关者参与以及持续监测既定性能基准的全面上市后监测。
  • 应建立管理差异和不良事件报告的清晰流程以维持信任与安全。
  • 强调为所有医疗专业人员提供包含儿科安全考量的基础AI素养课程,并为直接参与儿科影像的人员提供专门培训。
  • 公众与患者参与对于促进对儿科放射学中AI的理解和接受至关重要。

局限性

  • 原文未报告局限性

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

Artificial intelligence (AI) has potential to revolutionize radiology, yet current solutions and guidelines are predominantly focused on adult populations, often overlooking the specific requirements of children. This is important because children differ significantly from adults in terms of physiology, developmental stages, and clinical needs, necessitating tailored approaches for the safe and effective integration of AI tools. This multisociety position statement systematically addresses four critical pillars of AI adoption: (1) regulation and purchasing, (2) implementation and integration, (3) interpretation and postmarket surveillance, and (4) education. We propose pediatric-specific safety ratings, inclusion of datasets from diverse pediatric populations, quantifiable transparency metrics, and explainability of models to mitigate biases and ensure AI systems are appropriate for use in children. Risk assessment, dataset diversity, transparency, and cybersecurity are important steps in regulation and purchasing. For successful implementation, a phased strategy is recommended, involving early pilot testing, stakeholder engagement, and comprehensive postmarket surveillance with continuous monitoring of defined performance benchmarks. Clear protocols for managing discrepancies and adverse incident reporting are essential to maintain trust and safety. Moreover, we emphasize the need for foundational AI literacy courses for all health care professionals that include pediatric safety considerations, alongside specialized training for those directly involved in pediatric imaging. Public and patient engagement is crucial to foster understanding and acceptance of AI in pediatric radiology. Ultimately, we advocate for a child-centered framework for AI integration, ensuring that the distinct needs of children are prioritized and that their safety, accuracy, and overall well-being are safeguarded.

来源:J Am Coll Radiol,查看原文。版权归原作者所有。

引用

  1. Shelmerdine SC, Naidoo J, Kelly BS, et al. Artificial Intelligence Implementation in Pediatric Radiology for Patient Safety: A Multisociety Statement From the ACR, ESPR, SPR, SLARP, AOSPR, SPIN. J Am Coll Radiol. 2025. PMID: 41288534. doi:10.1016/j.jacr.2025.08.019

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。