Deep learning model opportunistically screens for osteoporosis on chest X-rays

Deep learning model using chest X-rays and clinical data achieved 85.8% accuracy (95% CI 78.5–90.9%) for osteoporosis screening in internal test set (AUC 0.95); external validation required.

• The multi-modal model achieved AUC-macro 0.95, with sensitivities of 84.2% for normal, 90.2% for osteopenia, and 82.3% for osteoporosis.
• Inclusion of DEXA wrist measurements reclassified 13.8% of patients, often upgrading severity, underscoring the value of cortical bone sites.
• Findings are from single-centre internal validation; external prospective studies are required for clinical translation.

Automated summary

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