Can AI catch breast cancer earlier? Lancet study shows gains in detection without replacing radiologists

Artificial intelligence could soon play a larger role in breast cancer screening, after a large Swedish trial found that AI-supported mammography detected more cancers early and reduced the incidence of aggressive cancers diagnosed between routine screenings.

The findings come from the MASAI study, published recently in The Lancet, which analysed routine mammography data from nearly 100,000 women screened between April 2021 and December 2022 across Sweden.

Researchers compared standard screening where each mammogram is independently read by two radiologists with an AI-supported approach designed to assist, not replace, human expertise.

How AI changed mammogram reading

In the AI-supported group, an artificial intelligence system first analysed mammogram images and assessed cancer risk. Scans flagged as low-risk were reviewed by a single radiologist, while higher-risk images were double-read by two radiologists, with the AI highlighting suspicious regions.

In the standard group, all scans were reviewed independently by two radiologists without AI assistance.

Earlier detection, fewer aggressive cancers

The study found that about 81% of breast cancers in the AI-supported group were detected during routine screening, compared with 74% in the standard group. Earlier detection is crucial, as cancers found during screening are typically smaller, less advanced and easier to treat.

Researchers also observed a notable reduction in aggressive cancer subtypes. The AI-supported group recorded 27% fewer aggressive cancers diagnosed later often referred to as interval cancers suggesting that AI may help identify fast-growing tumours before they progress.

Interval cancers, which are diagnosed between scheduled screening rounds because they were either missed or developed rapidly, were consistently lower in the AI-supported group than in standard screening.

AI as assistant, not replacement

Crucially, the researchers stressed that AI does not replace radiologists. Instead, it acts as a decision-support tool, helping prioritise cases and draw attention to potentially concerning findings.

Radiologists remained responsible for final diagnoses, with AI functioning like a highly attentive assistant fast, consistent and tireless, but dependent on clinical judgement.

Caution before global rollout

While the results are encouraging, experts caution that broader evidence is needed. Breast cancer screening programmes vary significantly across countries in terms of technology, workforce training and follow-up care.

Further studies across different populations and healthcare systems will be required to determine whether similar benefits can be replicated beyond Sweden.

Still, the findings suggest that AI-assisted screening could represent a meaningful shift in how mammograms are read enhancing early detection while keeping human expertise firmly at the centre of care.