文献速递儿科影像技术与安全

儿童放射学中AI整合:国际学术领袖视角

AI integration in pediatric radiology: perspectives from international academic leaders.
PubMed 文献 · Eur Radiol · 2026-07-04

骨龄评估是应用最广泛的工具(44.4%)。主要障碍为儿童专用数据集缺乏(83.3%)和整合困难(66.7%),关键促进因素是供应商成熟度与整合能力(72.2%)和内部推动者(66.7%)。

要点

  • 骨龄评估仍是儿童放射AI最普及的临床应用(44.4%),可作为科室启动AI项目的首选切入点。
  • 儿童专用数据集缺乏是最大障碍(83.3%),对AI输出的儿童适用性需保持审慎判断。
  • 供应商的整合成熟度(72.2%)比单纯算法性能更影响日常落地,采购时需重点考察工作流集成能力。

核心结论

  • 全球儿科放射学领导者对AI采用持谨慎但增长的态度,多数中心已部署至少一种AI工具,但感知影响仍属中等。
  • 缺乏儿科专用数据集、整合挑战、高成本或ROI不明确是主要障碍;人工因素而非技术因素是主要促进因素。
  • 55.6%的领导者认为当前儿科AI研究过度强调模型开发而忽视临床整合、伦理和可持续性。

研究设计与样本

  • 国际横断面调查研究,研究对象为全球儿童放射中心的科室主任和部门负责人。
  • 分析方法为描述性统计,未涉及影像设备或序列层面的技术参数比较。

关键影像表现/征象

  • 骨年龄评估是最广泛应用的AI工具(44.4%),其次为图像分割与量化(22.2%)、成像方案优化(16.7%)和自然语言处理(16.7%)。
  • 仅16.7%的中心将AI描述为「转型性」影响,平均临床影响评分为3.56/5,提示AI尚未根本改变儿科影像工作流程。
  • NLP和报告相关应用在北美地区聚集,转型性影响评分集中在亚太中心,提示地区间AI应用方向存在差异。

诊断标准与关键数值

  • 平均临床影响评分为3.56/5,仅16.7%的中心将AI评为「转型性」影响等级。
  • 最常见障碍:缺乏儿科专用数据集(83.3%)、整合挑战(66.7%)、高成本或不明确的ROI(50%)、网络安全问题(44.4%)。
  • 最常见促进因素:供应商成熟度与整合能力(72.2%)、内部推动者(66.7%)。
  • 55.6%的受访者同意当前儿科AI研究过度强调模型开发而忽视临床整合、伦理和可持续性。

临床应用与阅片要点

  • 评估AI工具时应关注其在儿科人群中的验证证据,不应直接套用成人数据集的性能指标到儿童影像判读。
  • 骨年龄评估AI可作为儿科放射科室验证AI部署流程和评估ROI的优先场景,基于其相对成熟的落地条件。
  • 推进AI项目时需识别并授权内部推动者,其在科室层面的倡导对成功整合的贡献不低于技术成熟度本身。
  • 将AI纳入儿科报告流程时,需关注网络安全和未成年人数据合规要求,44.4%的中心将网络安全列为障碍。

局限性

  • 横断面调查设计仅反映特定时间点的领导者主观感知,缺乏纵向数据追踪AI采用随时间的演变。
  • 基于自报问卷的数据可能存在报告偏倚,实际AI使用频率与临床影响评分不能等同于客观效果验证。
  • 应答率为69%,无应答的中心可能存在系统性特征差异,影响外推范围。

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

OBJECTIVE: Artificial intelligence (AI) is increasingly integrated into radiology, but pediatric imaging remains underrepresented in implementation studies. To assess the current status, barriers, and enablers of AI adoption in Pediatric Radiology from a global leadership perspective.

MATERIALS AND METHODS: A cross-sectional international survey of department leaders and division chiefs from pediatric radiology centers worldwide was conducted. The questionnaire included 14 items across domains of AI deployment, enablers, barriers, ethical concerns, stakeholder involvement, and future directions. Descriptive statistics were used for analysis.

RESULTS: Eighteen institutions completed the survey (69% response rate). Sixteen centers (88.9%) reported implementing at least one AI tool, with bone age assessment remaining the most widely used application (44.4%). Other applications included image segmentation and quantification (22.2%), imaging protocol optimization (16.7%), and natural language processing (16.7%). The mean clinical impact rating was 3.56/5, with only 16.7% describing AI as "transformational." The most frequently cited barriers were lack of pediatric-specific datasets (83.3%), integration challenges (66.7%), high cost or unclear return on investment (ROI) (50%), and cybersecurity concerns (44.4%). Enablers were primarily human rather than technical, with vendor maturity and integration (72.2%) and internal champions (66.7%) most frequently highlighted. Regional variation was observed, with NLP/reporting clustered in North America and transformational impact ratings concentrated in the Asia-Pacific centers. A slim majority (55.6%) agreed that pediatric AI research currently overemphasizes model development over clinical integration, ethics, and sustainability.

CONCLUSION: Global pediatric radiology leaders report cautious but growing AI adoption. Addressing pediatric data scarcity, enhancing multicenter collaboration, and prioritizing clinical integration and sustainability will be essential for safe and effective deployment.

KEY POINTS: Question How are leading pediatric radiology centers worldwide implementing AI, and what institutional barriers and enablers determine successful clinical integration? Findings Most centers deploy at least one AI tool, but perceived impact remains modest due to pediatric data scarcity and workflow integration challenges. Clinical relevance Addressing pediatric-specific dataset limitations through multicenter collaboration, empowering institutional AI champions, and prioritizing clinical integration over model development alone will be essential for ensuring AI benefits children as effectively as adults.

来源:Eur Radiol,查看原文。版权归原作者所有。

引用

  1. Shah, Kelly, Palasis, et al. AI integration in pediatric radiology: perspectives from international academic leaders. Eur Radiol. 2026. PMID: 42401764. doi:10.1007/s00330-026-12728-9

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。