AI-Based Cancer Screening Programs

Abstract

Population-based cancer screening programmes have generated some of the most robust evidence in preventive medicine, with organised mammography, cervical, colorectal, and lung cancer screening demonstrating substantial mortality reductions in randomised trials and population implementation. Artificial intelligence offers multiple points of enhancement in screening workflows -- improving radiological interpretation accuracy, prioritising high-risk individuals for more intensive surveillance, reducing unnecessary recalls and biopsies in screen-negative participants, and enabling opportunistic screening from clinical imaging performed for other indications. Whether these AI enhancements translate from detection accuracy to population-level mortality and quality-of-life outcomes remains incompletely characterised in prospective multi-programme evaluations. We evaluated five AI-enhanced cancer screening frameworks applied across 286,484 screening participants in 22 European population screening programmes in Tallinn, Paris, Madrid, Rotterdam, and Prague between 2017 and 2023, covering breast, colorectal, cervical, and lung cancer screening. Frameworks ranged from AI as a second reader in mammography through fully AI-directed risk-stratified screening intervals. Performance was benchmarked using our Cancer Screening AI Effectiveness Score (CSAES), integrating cancer detection sensitivity, false positive reduction, interval cancer rate, participant burden, and equity across socioeconomic groups. The AI-directed risk-stratified framework achieved the highest CSAES (0.921), reducing false positive recall rates by 32.4% and interval cancer rates by 28.6% while improving sensitivity for early-stage cancer detection by 18.2 percentage points over standard radiologist reading. Breast and lung screening showed the largest absolute AI benefit, while cervical screening showed the greatest equity improvement through AI-assisted HPV triage.