Purpose <p>Fractures are significant complications linked to type 2 diabetes mellitus (T2DM). In this study, we investigated the role of a computed tomography (CT)-based bone radiomic model for predicting vertebral fractures (VFs) in older patients with T2DM.</p> Methods <p>A cohort of 1486 T2DM patients over the age of 50 years who were free from VFs from 2019 to 2021 were included and subsequently followed up until 2024. A total of 135 new cases of VFs were identified. A control group of 270 individuals without VFs during the follow-up was selected. Radiomic features were gathered from the main domains of the CT scans.</p> Results <p>Radiomics score (Radscore) (adjusted risk ratio (aHR) = 12.04; 95% confidence interval (CI): 5.88–24.64) was independently associated with VFs in T2DM patients. The radiomic model had an acceptable performance in predicting VF risk (area under the curve = 0.752 (95% CI: 0.712–0.792) for 2-year risk, 0.765 (95% CI: 0.714–0.816) for 3-year risk, and 0.811 (95% CI: 0.727–0.894) for 4-year risk). A combined model of the Radscore and bone CT attenuation demonstrated a greater AUC (AUC = 0.795 (95% CI: 0.747–0.843) for 2-year risk; 0.789 (95% CI: 0.730–0.837) for 3-year risk; and 0.842 (95% CI: 0.757–0.927) for 4-year risk) than models based on the Radscore or bone CT attenuation alone. The calibration and decision curves demonstrated the excellent predictive accuracy of the model.</p> Conclusions <p>Bone radiomics has great potential in predicting VFs in older patients with T2DM.</p>