DXA Foundation Model

A domain-informed self-supervised learning embedding model consolidating four longitudinal NIH datasets of dual-energy X-ray absorptiometry (DXA) scans.

The model outperforms the clinical-standard WHO hip fracture risk assessment tool by 10% and supports downstream tasks including mortality risk prediction, cardiovascular health monitoring, and metabolic disease assessment. Optimized for fairness metrics and stable subgroup performance.

Published in Communications Medicine (2022); foundation model paper in preparation for Radiology.