The aim of this study was to develop and validate a machine learning–based model that incorporates multi-modal radiomics features from preoperative MRI and CT imaging, as well as clinical variables, to help predict vertebral refracture risk after percutaneous kyphoplasty (PKP) in post-menopausal women with osteoporotic vertebral compression fractures (OVCFs). A total of 156 patients (35 with refracture; 121 without refracture) were included, and the authors extracted over 3,600 radiomic features from preoperative lumbar T2-weighted MRI and CT images, which characterize important measurements such as bone density, texture, and trabecular microarchitecture. Radiomic features obtained from imaging modalities were combined with meaningful clinical data (e.g. age, BMI, vertebral Hounsfield units (HU), and comorbidity history) to generate and compare predictive models that used radiomic, clinical, and combined data sources. Among the models tested, the support vector machine (SVM) algorithm using radiomics features had an area under the curve (AUC) of 0.798 and the K-nearest neighbors (KNN) clinical model had an AUC of 0.74. The combined model using radiomics and clinical outcome achieved the best performance (AUC = 0.886), indicating that the combination of root quantitative features obtained from imaging along with traditional clinical factors will improve predictive accuracy. Patients who had a refracture were older and had lower vertebral bone density (97.0 ± 6.3 vs. 102.5 ± 4.7 HU, P < 0.001). The study demonstrates that MRI and CT provide complementary information: CT features captured trabecular features including density and external structure, while MRI showed marrow composition and changes in hydration associated with bone fragility.