AI-Assisted Mammography Studies Show Higher Detection of Aggressive Breast Cancers

Recent research from multiple countries has examined how artificial intelligence (AI) can support breast cancer screening by improving the detection of aggressive cancers that may be missed between regular mammograms.

Interval breast cancers, which are diagnosed between scheduled screening rounds, are often associated with poorer outcomes compared to cancers detected during routine screening. These interval cancers can include missed-reading errors, minimal signs, occult cancers, and true interval cancers that were not visible at the time of screening.

In Sweden, the MASAI trial involved over 100,000 women and compared AI-assisted mammography with the standard double reading approach. The trial found that 81% of cancers in the AI group were detected during screening, compared to 74% in the control group, without an increase in false positive results.

The MASAI trial also reported a reduction in the number of aggressive or advanced cancers diagnosed between screening rounds when AI-supported screening was used. The AI system in this trial assigned risk levels to mammograms, directing low-risk cases to single reading and higher-risk cases to double reading, while highlighting suspicious areas for further review.

What the numbers show

• In the UK, an AI tool identified up to 42.4% of interval cancers by flagging the top 20% of risk scores
• A UCLA study found AI flagged 90% of missed-reading error interval cancers in nearly 185,000 mammograms
• German research showed AI detected and localised 27.5% of false-negative and minimal-signs interval cancers at 99% specificity

A retrospective study in the UK used an AI tool called Mirai to assign risk scores to over 134,000 screening mammograms. This study found that the AI could identify up to 42.4% of interval cancers by focusing on the 20% of women with the highest risk scores, resulting in an additional detection rate of 1.7 per 1,000 screened women.

The UCLA-led study in the United States analyzed nearly 185,000 mammograms and found that AI flagged 90% of interval cancers that were missed due to reading errors, as well as high percentages of other interval cancer types. The study estimated that using AI in screening could reduce mammographically visible interval cancers by about 30%.

German researchers conducted a retrospective analysis covering over a decade of screening data. They found that, at a specificity of 99%, AI could detect and correctly localise 27.5% of false-negative and minimal-signs interval cancers, and 21.1% of advanced or metastatic cases that were retrospectively visible. The same study reported that AI correctly localised 2.8% of true interval cancers, which are those not visible at prior screening.

Across these studies, the use of AI in mammography screening has been associated with higher detection rates of interval cancers, particularly those that are more aggressive or advanced. The findings suggest that AI tools may help identify cancers that are otherwise missed between routine screenings, without increasing false positives according to trial data.

* This article is based on publicly available information at the time of writing.