Explainable opportunistic osteoporosis screening from chest X-rays: a retrospective comparison of foundation models
Authors' conclusion
Our findings underscore the necessity of incorporating explainability as a key criterion when selecting deep learning models for opportunistic osteoporosis screening.
Study context
- Population
- Female patients with chest X-rays paired with BMD scores, Seoul National University Hospital21,031 chest X-rays
- Participants
- 14,502 patients
- Comparison
- Twelve foundation model variants (natural- and medical-domain models, various fine-tuning strategies) for classifying osteoporosis status from chest X-rays
- Outcome
- Classification of osteoporosis status (normal, osteopenia or osteoporosis): predictive performance (AUC, accuracy, sensitivity, specificity) and explainability (occlusion and saliency-map analysis)
Methodological detailsretrospective, comparison of foundation models, single-centre
- Retrospective
- Comparison of foundation models
- Single-centre
Performance
As reported in the abstract
Predictive performance (highest among models)
| Test or model | AUC | AUC 95% CI |
|---|---|---|
| DINOv2 with low-rank adaptation | 0.93 | 0.92-0.94 |
DINOv2 with low-rank adaptation
AUC0.93
AUC 95% CI0.92-0.94
Further findings
- Medical foundation models did not consistently outperform natural-domain models
- Higher performance did not always correlate with better explainability
Original publication
Explainable opportunistic osteoporosis screening from chest X-rays: a retrospective comparison of foundation models. Osteoporos Int. Published 27 October 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →