29 posters, 13 topics, 90 authors, 29 institutions
ePostersLive by SciGen Technologies S.A. All rights reserved.
Clinical Radiology Audit and Quality Improvement ePoster Competition 2025
18 NOVEMBER, 2025 | LONDON, UK
48
Improving access to CT for suspected lung cancer with Artifical intelligence.
Poster Presenter
Part of Topic
AI
Introduction
Lung cancer is the leading cause of cancer-related death in the United Kingdom, where 5-year survival rates are among the lowest compared with similarly developed nations1. The National Optimised Lung Cancer Pathway (NOLCP) seeks to address this inequality with two key diagnostic goals to improve outcomes:
Rapid diagnosis(within 3 days) andEarly-stage diagnosis(Stage I–II), when curative treatments are most effective2. Artificial Intelligence (AI) shows early promise in supporting these goals, but real-world validation is critically lacking3. Our project represents the largest real-world study to date investigating the use of a commercial product, Annalise CXR Enterprise V2.34, in addressing these national targets.
Methods
Prospective pre-post study over 3 years, comparing NOLCP target outcomes (time to diagnosis and stage at diagnosis) with and without AI triage of consecutive GP-referred chest X-rays (CXRs) showing features of cancer
Discussion
Rapid diagnosis may improve early access to treatment and subsequent survival5. However, this surrogate outcome is confounded by concurrent service developments and relies on several assumptions to demonstrate survival benefit. Contrary to prior claims and reports6, implementation of AI triage did not increase the detection of early-stage cancers, likely reflecting the well-established limitations of chest radiography in this task7. Without an impact on early-stage diagnosis, AI triage of CXRs alone is unlikely to meet cost-effectiveness thresholds for lung cancer. Other use cases—such as nasogastric tube detection, critical finding alerts, and exclusion of normal radiographs—may offer additional utility. High-risk screening programmes and improved access to CT imaging are likely more effective avenues for improving lung cancer outcomes.
Recommendations
Several lessons can be taken from this AI deployment to inform future use:
-
Clinical experts must lead AI deployment within bespoke pathways with narrow outcomes and supported by robust scientific evidence.
-
Post-deployment surveillance is paramount using clinically meaningful outcome metrics and avoiding the biased review of only positive cases
. -
Pay-for-outcome schemes are powerful motivators linking procurement to performance. Pay for product schemes with vague objectives are inadequate and . instead, incentives such as additional payment for the detection of early-stage cancer, when used within an appropriate framework, incentivise vendors to provide the best possible outcomes for patients.