Breast screening: AI triage removes 39,834 readings in Córdoba, increases detection by 15.2%

Out of 31,301 women, a prospective Spanish trial shows a 63.6% reduction in reading load and more cancers detected, at the cost of an increased recall rate.

Between March 2022 and January 2024, Córdoba's breast cancer screening unit compared, of the same 31,301 women, classical dual radiological reading and a strategy where artificial intelligence sorts examinations before any human intervention. The trial, prospective and designed to establish the non-inferiority of the second approach, was published in Nature Medicine.

Automated sorting eliminated 39,834 readings, representing a 63.6% decrease in the workload of radiologists. On detection, the strategy did not simply equal double reading: it statistically exceeded it, with a rate of cancer detected higher than 15.2%.

The counterpart is on the callback side. The rate of convocation for further examinations is increasing by 14.8%, and that of false positives is also increasing. The authors therefore insist on careful implementation: continuous quality assurance, post-deployment surveillance and validation in various screening settings, especially since some cancers were classified as low risk by the system.