In the United States, 1 in 8 women are diagnosed with breast cancer. Breast tumor heterogeneity is well-established, with intratumor heterogeneity manifesting spatially and temporally. Increased heterogeneity is associated with adverse clinical outcomes. Current critical disease treatment decisions are made on the basis of biomarkers acquired from tissue samples, largely under sampling the heterogeneous disease burden. In order to drive precision medicine treatment strategies for cancer, personalized biomarkers are needed to truly characterize intratumor heterogeneity. Medical imaging can provide anon-invasive, whole tumor sampling of disease burden at the time of diagnosis and allows for longitudinal monitoring of disease progression. The studies outlined in this thesis introduce analytical tools developed through computer vision, bioinformatics, and machine learning and use diagnostic and longitudinal clinical images of breast cancer to develop computational imaging biomarkers characterizing intratumor heterogeneity. Intrinsic imaging phenotypes of spatial heterogeneity, identified in dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) images at the time of diagnosis, were identified and validated, demonstrating improved prognostic value over conventional histopathologic biomarkers when predicting 10-year recurrence free survival. Intrinsic phenotypes of longitudinal change in spatial heterogeneity in response to neoadjuvant treatment, identified in DCE-MRI were identified and leveraged as prognostic and predictive biomarkers, demonstrating augmented prognostic value when added to conventional histopathologic and personalized molecular biomarkers. To better characterize 4-D spatial and temporal heterogeneity, illuminated through dynamic positron emission tomography imaging, a novel 4-D segmentation algorithm was developed to identify spatially constrained, functionally discrete intratumor sub-regions. Quantifying the identified sub-regions through a novel imaging signature demonstrated the prognostic value of characterizing intratumor heterogeneity when predicting recurrence free survival, demonstrating prognostic improvement over established histopathologic biomarkers and conventional kinetic model derived parameters. Collectively, the studies in this thesis demonstrate the value of leveraging computational imaging biomarkers to characterize intratumor heterogeneity. Such biomarkers have the potential to be utilized towards precision medicine for cancer care.