Recent study evaluated the AI deep-learning model Mirai for predicting interval breast cancers (ICs)—cancers that appear between scheduled mammogram screenings—in a screening program for women aged 50–70.
The study analyzed 134,217 negative screening mammograms, including 524 ICs. Mirai’s 3-year risk scores were significantly higher in women who later developed ICs, with an overall C index of 0.70. Yearly prediction performance remained consistent (AUCs: 0.72 for year 1, 0.67 for years 2 and 3). Performance was similar across age groups and breast densities, suggesting Mirai naturally accounts for density as a risk factor.
Using a practical threshold, identifying the top 20% of women by Mirai risk scores would have detected 42.4% of ICs, equating to 1.7 extra cancers per 1,000 women screened.
The study concludes that Mirai could help identify women at higher risk of ICs, supporting decisions on supplemental imaging or shortened screening intervals to catch cancers earlier.