Multicenter AI Fuses MRI Muscle Features and BMD to Predict Osteoporotic Vertebral Fractures

An interpretable AI model fusing MRI paraspinal muscle features with BMD and clinical data achieved AUCs of 0.881–0.917 for osteoporotic vertebral fracture prediction across multiple centers, significantly outperforming a BMD-only baseline.

• Compared with the clinical-BMD baseline, the model improved AUC by 0.046–0.093 (DeLong P≤0.034 across most cohorts).
• Continuous net reclassification improvement (NRI) ranged from 0.268 to 1.034 across validation cohorts (P≤0.011).
• MSF-OVF significantly outperformed single-muscle models (AUC 0.454–0.736, P≤0.009).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.