The MASAI trial, led by Kristina Lång, associate professor in diagnostic radiology at Lund University, is the first randomised trial of its scale to test artificial intelligence in population breast screening. Some 106,000 Swedish women aged 40 to 74 were randomised: in one arm, AI triaged mammograms to single or double reading based on risk scores and flagged suspicious findings, while the control arm received standard double reading by two breast radiologists.
The final results, published in The Lancet, focus on interval cancers, the tumours that surface between screening rounds and tend to be the most aggressive. AI-supported screening produced 12 per cent fewer interval cancers overall, 82 cases against 93, with 16 per cent fewer invasive interval cancers, 19 per cent fewer large ones, and 27 per cent fewer aggressive subtypes. Earlier results had already shown 29 per cent more cancers detected at screening, mostly small and lymph-node negative, with sensitivity rising without an increase in false positives.
The workload finding may prove as consequential as the clinical one: radiologist screen-reading effort fell by 44 per cent, in a specialty facing chronic staff shortages across Europe. Several Swedish regions have already begun implementing AI-supported screening, and the researchers describe integration as relatively straightforward, largely a matter of adding AI software to existing systems. For screening programmes internationally, MASAI supplies the randomised evidence that cautious health systems have been waiting for.