A 105,000-woman trial shows AI can raise cancer detection and halve reading workload

Screening programmes are where medical AI meets the largest number of healthy people, so the evidence bar is high. The MASAI trial cleared it.

MASAI is a randomised, controlled, population-based screening trial of more than 105,000 women in Sweden, comparing AI-supported mammography reading against the standard of two radiologists reading independently. Final results were published in The Lancet. ¹²

What AI changed

• Cancer detection rate rose by 29 percent with AI support compared with standard double reading. ²
• Screen-reading workload fell by 44 percent, because AI triaged which cases needed one versus two human readers. ²
• The interval cancer rate, cancers that surface between screening rounds and are the ones you most want to catch, was 12 percent lower in the AI-supported group, without an increase in false positives. ¹

Crucially, the AI-supported arm did not trade detection for a flood of unnecessary recalls: sensitivity went up while specificity held. ¹

Reading it carefully

This is a single-country trial within one national screening programme, and long-term outcomes such as mortality are still to come. AI here is a decision-support layer inside a mature double-reading system, not a replacement for radiologists.

But as a piece of evidence it is unusually robust: randomised, prospective, and large. For health systems weighing whether AI belongs in screening, MASAI moves the question from "is it safe" toward "how do we deploy it responsibly at scale."