New model-driven imaging research aims to shift cancer detection and staging earlier in the care pathway. A preoperative AI approach using MRI radiomics and clinical variables predicted lymphovascular invasion risk in gastric cancer before surgery, with the goal of stratifying patients by disease-free survival risk. In breast cancer, another multicenter deep learning model targeted microcalcification interpretation, reporting high-accuracy prediction for malignancy in BI-RADS 4 cases and potential reductions in unnecessary biopsies. Together, these updates reflect an ongoing push toward decision-support tools that reduce overtreatment by improving risk stratification—an area where clinical utility depends on robust validation beyond single-center datasets.