An important goal of biomedical research has been the prediction of cancer recurrences at the earliest possible times. We now review the first method to accurately predict the outcomes of ductal carcinoma in situ (DCIS) of the breast patients. The method combines the immunofluorescent localization of enzymes and transporters participating in heightened glucose uptake with machine learning-based computer vision to predict patient outcomes. Recurrent breast cancer is preceded by changes of the intracellular positions of phospho-Ser226-glucose transporter type 1, phosphofructokinase type L, and phosphofructokinase/fructose 2,6-bisphosphatase type 4 within ductal epithelial cells of DCIS lesions, as judged by monoclonal antibody staining. These enzymes form peripheral enzyme clusters that increase kinetics and amounts of product formation due to enzyme proximity. However, these same enzymes are not recruited to the plasma membrane’s vicinity in epithelial cells of lesions of nonrecurrent patients. Intracellular enzyme trafficking to the cell periphery will also promote NADPH and GSH formation. Assembly of the machinery to perform a cancer recurrence is likely to be incomplete within very early lesions and may not be recognizable to pathologists. To reduce the impact of this issue, artificial intelligence/machine learning was used to identify 1) intermediate structures between the recurrent and nonrecurrent phenotypes and 2) stromal biomarker patterns correlating with recurrent or nonrecurrent disease. All image features are contained within software models to predict patient outcomes. Using the phospho-Ser226-glucose transporter type 1, phosphofructokinase type L, and phosphofructokinase/fructose 2,6-bisphosphatase type 4 biomarkers to evaluate patient samples, high levels of accuracy (91.5%) are obtained with no false negatives for all DCIS lesions. We propose that cancer recurrences arise from intracellular enzyme trafficking following desequestration of enzymes from nucleoli of ductal epithelial cells.