A new machine-learning model can identify the stage of disease in ductal carcinoma in situ, a type of preinvasive tumor that can sometimes progress to a deadly form of breast cancer. This could help clinicians avoid overtreating patients whose disease is unlikely to progress.