AI赋能影像筛查的规模化实施:英格兰肺癌筛查经验
本文总结 NHS 英格兰肺癌筛查计划的技术与人力架构。核心技术要素包括云原生影像 IT、虚拟国家专科网络及 AI 结节检测的整合。
要点
- 分布式采集、集中式专家判读与标准化结构化报告是同质化筛查的关键架构。
- 专科放射科医生虚拟网络优于事务性远程放射学合同,可实现同行评议与问责。
- 供应商中立 AI 结节检测支持网络级漂移监测与人群偏倚评估。
核心结论
- 肺癌筛查可视为影像筛查规模化的早期原型,而非特例。
- 其架构可适应性推广至前列腺 MRI、CT 结肠成像筛查及未来潜在的心脏 CT 筛查。
研究设计与样本
- 技术支持:云原生、供应商中立的影像信息技术;自 2020 年起支持超一百万例研究。
- 人力支持:虚拟国家专科网络,含 180 名具有胸部亚专科背景的顾问放射科医生。
关键影像表现/征象
- AI 结节检测/容积测量整合用于网络级部署后监测。
- 当前每月扫描超 41000 次,超 99% 在 72 小时内出具报告。
诊断标准与关键数值
- 检出肺癌 7193 例,63.1% 在 I 期检出。
- 报告时效标准为 72 小时内返回,实现率超 99%。
临床应用与阅片要点
- 应构建云原生且供应商中立的影像 IT 平台,以克服不同 PACS 系统分散问题。
- 采用单一报告环境内的专科网络工作模式,而非事务性远程报告。
- 将 AI 工具纳入监测流程,关注算法漂移和人群偏倚。
局限性
- 不同筛查路径在证据基础、邀请模式和阅片要求上存在差异。
- 部分应用仍处于研究阶段。
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
Lung cancer is the leading cause of cancer death globally, and low-dose computed tomography (LDCT) screening reduces lung cancer mortality. The NHS England Lung Cancer Screening Programme is the largest national LDCT implementation to date, with over 2.5 million invitations issued, 7,193 cancers diagnosed and 63.1% detected at stage I in its first five-year evaluation. We describe the technical and workforce architecture supporting interpretation at scale and its transferability to other imaging-based screening. Three components have proved central. First, cloud-native, vendor-agnostic imaging information technology has overcome the fragmentation of NHS picture archiving and communication systems, supporting distributed acquisition, centralised expert interpretation and standardised structured reporting. Second, a virtual national specialist network of 180 consultant radiologists with thoracic subspecialty interest working within a single reporting environment, rather than a transactional teleradiology contract, has enabled subspecialist delivery, embedded peer review, discrepancy logging and named-consultant accountability. Third, vendor-agnostic integration of artificial-intelligence nodule detection/volumetry has supported network-level post-deployment monitoring for algorithm drift and population bias. This has supported over one million studies since 2020, currently exceeding 41,000 scans per month, with >99% returned within 72 hours. Established breast screening already exemplifies these principles: organised double reading, centralised quality assurance and, increasingly, integrated AI. With programme-specific adaptation, the same approach is relevant to screening with prostate MRI, CT colonography, and potentially cardiac CT in the future. Although these pathways differ in evidence base, invitation model and reading requirements and some applications remain investigational. Lung cancer screening may therefore represent an early archetype, rather than a special case, for imaging-based screening.
引用
- Hare, Robinson, Tavare, et al. Scaling AI-enabled imaging-based screening: lessons from reporting for the NHS England lung cancer screening programme. Clin Radiol. 2026. PMID: 42567112. doi:10.1016/j.crad.2026.107429
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。