Breast screening
A Swedish trial of 106,000 women: 29% more cancers found, 44% less radiologist workload, 12% fewer interval cancers, and among them 27% fewer aggressive subtypes (The Lancet, January 2026). Scotland: 12% more cancers among 10,889 women, and in a workflow model the workload fell by up to 30% (2024). The United Kingdom started a trial with nearly 70,000 women at 30 sites and 11 million pounds of funding (2025). In Russia, AI mammography is among the studies paid for by compulsory health insurance; in the Krasnodar Territory, a case is described of an aggressive tumor that AI found after a clean first reading.
Other cancer sites
Pancreatic cancer: a model on routine CT without contrast found 31 cases among more than 20,000 patients where radiologists saw nothing, with one false alarm per 1,000 (Nature Medicine, 2023). Colonoscopy: a Russian randomized trial with 1,128 patients raised adenoma detection from 41.3% to 47.2%, and after midday the effect grew to 9.2 points. Primary care: a risk platform in 1,084 practices in England, according to the developer, raised the share of confirmed cancers by 20% against the national average and avoided more than 61,000 unneeded urgent referrals.
What this means for clinics
The effect of AI in cancer care is measured by three numbers: extra cases found, unneeded referrals and doctor time saved. Rollout goes through registered medical devices and a second reading by a doctor. The service side of cancer care, such as booking, reminders for follow-up exams and documents, can be automated without registration.