Deep Learning-based Bone Mineral Density Prediction Using Pediatric Chest Radiographs: A Multicenter Feasibility...

Area of Science:

• Radiology
• Artificial Intelligence
• Pediatric Health

Background:

• Bone mineral density (BMD) assessment is crucial for pediatric bone health.
• Dual-energy x-ray absorptiometry (DXA) is the gold standard but lacks widespread accessibility.
• Accessible methods for pediatric BMD evaluation are needed.

Purpose of the Study:

• To develop and validate an AI model for predicting pediatric BMD from standard chest radiographs.
• To assess the model's performance in identifying low BMD in children.

Main Methods:

• A retrospective study utilized chest radiographs and clinical data (age, sex, height, weight) from 1464 pediatric patients.
• An AI model was trained and tested on internal and external datasets, predicting lumbar spine BMD Z-scores.
• Performance was measured using Pearson correlation for BMD prediction and AUC for low BMD detection.

Main Results:

• The AI model demonstrated strong correlations with DXA-derived BMD Z-scores in both internal (r=0.85) and external (r=0.76) test sets.
• The model achieved high accuracy in detecting low BMD, with AUCs of 0.92 (internal) and 0.90 (external).
• High sensitivity and specificity were observed in both test cohorts for identifying low BMD.

Conclusions:

• An AI model utilizing chest radiographs can accurately predict pediatric BMD Z-scores.
• This AI approach shows promise for identifying children with low BMD, improving accessibility to bone health assessments.