AI Algorithm Could Refine Breast Cancer Screening by Identifying High-Risk Cases Missed in Routine Exams

A study published in Radiology reveals that an artificial intelligence‑driven system may significantly enhance breast cancer screening by identifying women who are especially likely to develop so‑called interval cancers—those diagnosed between regular mammograms and often associated with worse outcomes.

The research team from the United Kingdom used screening data from over 134,000 women aged 50 to 70, collected between 2014 and 2016 at two centers in the national screening program. Of those women, 524 developed interval breast cancers. By applying the deep‑learning algorithm known as Mirai to previously negative screening mammograms, the investigators assessed whether the model could predict these later‑occurring cancers. The AI tool used image features such as breast density and subtle structural changes in the tissue.

The investigators reported that “interval cancers generally have a worse prognosis compared with screen‑detected cancers, because they tend to be either larger or more aggressive,” explained co‑author Fiona J. Gilbert, M.B.Ch.B., professor of radiology at the University of Cambridge and honorary consultant radiologist at Addenbrooke’s Hospital. “That’s why it’s important to minimize the number of interval cancers that you have in any screening program.”

Results showed that among women assigned the highest risk scores by the AI (top 20 % of scores), the algorithm retrospectively predicted 42.4 % of interval cancers. In smaller subsets, the top 10 % and top 5 % of scores captured 26.1 % and 14.5 % of interval cancers respectively, while the very highest 1 % captured 3.6 %. The researchers noted that these corresponded to additional cancer detection rates of 1.7, 1.0, 0.6 and 0.1 per 1,000 screened women in those risk strata. According to lead researcher Joshua W. D. Rothwell, the findings mean that “further workup of mammograms within the top 20 % of scores could yield 42.4 % of interval cancers, meaning that Mirai could be used to identify women for supplemental imaging or a shortened screening interval, instead of or in addition to breast density.”

In discussing the technology, Dr. Gilbert said “Personalized breast cancer screening depends on accurately assessing an individual’s risk of developing breast cancer within a specific timeframe. We can use supplemental imaging and adjust screening frequency based on a woman’s breast density and likelihood of developing breast cancer within a short timeframe.”

Notably, the tool performed best in predicting interval cancers that appeared within one year of the screening mammogram. Its performance declined somewhat for cancers occurring in the 12–24‑month or 24–36‑month windows. The algorithm was less effective in women with extremely dense breast tissue, but still outperformed conventional risk‑prediction tools.

In practical terms, if the system were used to recall the top 20 % of women for supplemental imaging, the researchers note that the UK’s program would need capacity to offer contrast‑enhanced mammography or MRI to roughly 440,000 women annually, given that approximately 2.2 million UK women are screened each year. “Identifying women at an increased risk of developing breast cancer is a complex, multifactorial problem,” Dr. Gilbert added. “The goal is to accurately identify the women most likely to have an interval cancer while minimizing the volume of supplemental imaging performed.”

The next steps outlined by the team include comparing different commercially available predictive tools, performing economic modelling and cost‑effectiveness analyses, and conducting a prospective trial to assess whether AI‑based risk stratification can be implemented in real screening settings.

With this evidence, AI‑enabled risk‑prediction tools like Mirai could mark a shift from a one‑size‑fits‑all approach toward tailored screening strategies—potentially catching more cancers earlier and reducing the burden of aggressive disease presentation.