Reading a screening mammogram · Contested (since 2023)

In August 2023 the interim safety analysis of the Swedish MASAI trial, on 80 033 women, reported 244 cancers found with software support against 203 by standard double reading, with 44 percent fewer screen readings.

MASAI did not remove the radiologist. It replaced one of the two readers that the Swedish protocol requires, kept a human on every case the software flagged, and left the recall decision to a radiologist. The final results, published in The Lancet in January 2026 on more than 100 000 women, report an interval cancer rate of 1.55 per 1000 against 1.76, fewer large and aggressive tumours, and a false positive rate of 1.5 percent against 1.4. The reading of the image is being handed over, and it is being handed over carefully, one reader at a time. Deciding to recall a woman, taking the biopsy and telling her the result were never part of what was tested.