AI From Screening Mammograms Flags Short‑Term Risk of Interval Breast Cancers

Overview

• In a retrospective study published in Radiology, Mirai analyzed 134,217 negative screening mammograms from two UK centers (2014–2016) to generate three‑year risk scores for interval cancers.
• The highest 1%, 5%, 10% and 20% of risk scores accounted for 3.6%, 14.5%, 26.1% and 42.4% of the 524 interval cancers, indicating a potential pathway for targeted supplemental imaging or shortened screening intervals.
• Prediction performance was strongest for cancers diagnosed within one year of screening and was reduced in women with extremely dense breast tissue.
• The model outperformed conventional risk tools in this cohort, though more than half of interval cancers were not captured within the highest‑risk groups.
• Researchers plan head‑to‑head comparisons with other AI tools, economic and cost‑effectiveness analyses, and prospective trials to evaluate clinical utility and feasibility.