Opportunistic MRI-based osteoporosis assessment in patients with lumbar vertebral compression fractures using vertebral radiomics and paraspinal muscle measurements

Routine lumbar spine MRI may provide an opportunity for opportunistic osteoporosis assessment in patients with vertebral compression fractures (VCFs). In this study, we focused on patients with lumbar VCFs and developed and externally validated MRI-based models combining vertebral radiomics and quantitative paraspinal muscle measures for osteoporosis identification. A retrospective study was conducted at two centers and enrolled 203 patients with lumbar VCFs who had undergone lumbar MRI and dual-energy X-ray absorptiometry (DXA) from 2023 to 2025. Patients from one center were assigned to the training cohort ( n = 85), whereas those from the other center formed the external validation cohort ( n = 118). Three models were built: a radiomics model from sagittal T1-weighted images, a clinical-muscle model based on clinical variables and quantitative paraspinal muscle measures from axial T2-weighted images, and a combined model. Model discrimination, calibration, and clinical usefulness were assessed using ROC analysis, calibration curves, and decision curve analysis; AUCs were compared using the DeLong test. Model interpretability was assessed using Shapley additive explanations (SHAP). Among 1,197 extracted radiomic features, 9 were retained. The clinical-muscle model included age and MF fat-related signal fraction after 2-pixel erosion (%), and the combined model further incorporated the radiomics score (Rad-score). In the external validation cohort, the combined model outperformed both the clinical-muscle and radiomics models. It achieved an area under the curve (AUC) of 0.967, compared with 0.914 for the clinical-muscle model and 0.829 for the radiomics model. All three models demonstrated acceptable calibration, and decision curve analysis (DCA) further indicated that the combined model provided additional clinical benefit. Shapley additive explanations (SHAP) suggested that both radiomic and paraspinal muscle variables contributed to model prediction. The combined MRI-based model performed best in opportunistic osteoporosis assessment for individuals with lumbar VCFs. Routine lumbar MRI may help identify patients who warrant formal osteoporosis evaluation in this setting. This approach should be regarded as a complementary triage tool to standard DXA and requires further validation before use in individuals without vertebral fractures.