AI breast cancer risk predictions remain stable over time and enable personalized screening intervals

AI-based 2-year breast cancer risk scores were stable for most women across consecutive mammograms, with a mean absolute change of just 0.006. For the 3.5% with a large score increase, short-term cancer incidence was 3.5-fold higher, supporting use of AI to guide personalized re…

• Retrospective cohort of 53,858 women with two consecutive screening mammograms analyzed with a validated deep-learning model; mean interval between exams was 2.1 years.
• Large AI risk-score increases (>0.20) occurred in only 3.5% of women but were associated with a 3.5-fold higher 2-year cancer incidence (95% CI 2.8–4.3) compared to women with stable scores.
• Single-institution, retrospective design; generalizability to other populations and AI models requires prospective, multi-site validation.

RadPigeon summaries are original and for information only. They are not clinical advice.