A Swedish trial is offering some of the clearest evidence yet that AI might sharpen, not replace, the human eye in breast cancer screening. In what researchers call the largest study of its kind, 100,000 women undergoing routine mammograms between April 2021 and December 2022 were randomly assigned to standard screening by two radiologists—a standard practice in Europe, per the CBC—or to AI-assisted screening. The AI software flagged suspicious areas and sorted images into low- and high-risk categories, with those in the latter category receiving double human reads as opposed to only one.
Results, published Thursday in the Lancet, suggest AI support meant more cancers were caught at the screening visit and fewer surfaced later. Some 81% of cancers in the AI group were found during screening, versus 74% without AI, while subsequent cancer diagnoses were 12% lower in the AI group when compared with the control. The AI group also saw 27% fewer aggressive cancer subtypes, per the Guardian.
Lead author Dr. Kristina Lång of Lund University said AI-backed mammography could both ease radiologist workload and push more cancers into the "caught early" category, when they are easier to treat, but emphasized the need for gradual, closely monitored rollout. The study does not recommend replacing clinicians: at least one radiologist was still required to read each scan. Outside experts called the findings encouraging but urged caution, noting they come from a single center and the race and ethnicities of the participants weren't logged. Researchers will next examine the cost of instituting AI-assisted screening in the real world, per the CBC.