Want to Know Your Future Breast-Cancer Risk? Just Ask AI (www.wsj.com)

Researchers and startups are developing AI models that predict a woman’s near‑term risk of developing breast cancer by analyzing historical mammograms — some from patients who later developed the disease. Trained on longitudinal imaging data, these algorithms learn subtle imaging patterns invisible to radiologists and output probabilistic risk estimates for the coming years. Similar approaches are being explored for lung cancer and other conditions, moving risk prediction from population-level screening to individualized forecasting. Technically, the work relies on deep‑learning models that pick up subvisual biomarkers and temporal changes in serial mammograms; key challenges include model calibration, prospective validation, and ensuring training datasets are demographically and scanner‑diverse to avoid bias. If robust and properly integrated, these tools could personalize screening intervals, prioritize high‑risk patients for earlier workups, and enable preventive interventions—potentially improving outcomes and resource allocation. But they also raise clinical concerns: false positives, overdiagnosis, patient anxiety, explainability, and regulatory and workflow hurdles. The promise is real, but broad clinical adoption will depend on transparent evaluation, external validation, and careful implementation.