Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
AI Model Predicts Bone Health from Chest X-Rays in Children
Researchers developed a deep learning model that predicts bone mineral density from pediatric chest radiographs with high accuracy, potentially offering a more accessible screening tool for pediatric bone health assessment.
Background
Bone mineral density assessment is critical for identifying children at skeletal risk, yet dual-energy x-ray absorptiometry (DXA)—the current reference standard—is underutilized in pediatric care due to limited accessibility. This multicenter feasibility study explored whether artificial intelligence could extract clinically relevant BMD information from routine chest radiographs, which are already widely performed.
Key Findings
- Strong correlation with DXA-measured BMD: r = 0.85 (internal test), r = 0.76 (external test; P < 0.001)
- Excellent diagnostic performance for detecting low BMD: AUC 0.92 internally, AUC 0.90 externally
- Successfully identified BMD abnormalities across diverse pediatric conditions including hemato-oncologic, inflammatory, renal, and neuromuscular disorders
- Short stature independently associated with higher prediction errors (OR 1.99)
Why It Matters
This chest radiograph-based AI approach enables opportunistic bone health screening using images already obtained for other clinical indications, without additional radiation exposure. By leveraging routine imaging, the model could improve early identification of children at skeletal risk and address critical underutilization of bone density screening in pediatric populations.
Limitations
The study involved 1,464 patients from two Korean tertiary hospitals. External validation showed moderately reduced performance compared to internal testing, suggesting the need for broader validation across diverse populations and healthcare settings before widespread clinical implementation.
Original paper: Deep Learning-based Bone Mineral Density Prediction Using Pediatric Chest Radiographs: A Multicenter Feasibility Study. — Radiology. 10.1148/radiol.252761
