Opportunistic screening for osteoporosis using chest X-rays and deep learning: A systematic review and meta-analysis.
Khaled Skaik, Jean Pierre Abdallah, Robert Koucheki et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated the diagnostic performance of deep learning algorithms applied to chest X-rays for osteoporosis detection, involving 200,796 patients across 12 studies. The pooled sensitivity was 0.85 and specificity was 0.83, indicating promising potential for clinical implementation despite high heterogeneity in results.
Key findings
- Pooled sensitivity of DL models was 0.85 [95% CI: 0.82-0.87].
- Pooled specificity of DL models was 0.83 [95% CI: 0.81-0.85].
- Summary AUC for all DL models was 0.69 [95% CI: 0.68-0.71].
- External validation AUCs ranged from 73.4% to 94.6%.
Why it matters
The findings suggest that deep learning models can enhance osteoporosis screening through chest X-rays, potentially improving early detection and management of this condition in clinical settings.
What to keep in mind
High heterogeneity in internal validation results and a limited number of externally validated studies were noted.