Scaling AI-enabled imaging-based screening: lessons from reporting for the NHS England lung cancer screening programme.
Authors
Affiliations (6)
Affiliations (6)
- The HLH Imaging Group LTD, Ruislip, Middlesex, UK; Department of Radiology, Royal Free London NHS Foundation Trust, London, UK.
- The HLH Imaging Group LTD, Ruislip, Middlesex, UK.
- The HLH Imaging Group LTD, Ruislip, Middlesex, UK; Department of Radiology, Northumbria Healthcare NHS Foundation Trust, North Shields, UK.
- Lung Cancer Screening Patient Citizen and Expert By Experience, UK.
- The HLH Imaging Group LTD, Ruislip, Middlesex, UK; Department of Radiology, Cambridge University Hospitals NHS Foundation Trust, Cambridge, UK.
- The HLH Imaging Group LTD, Ruislip, Middlesex, UK; Department of Radiology, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK. Electronic address: [email protected].
Abstract
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.