From Baseline to Follow-Up: Counterfactual Spine DXA Image Synthesis in UK Biobank Using a Causal Hierarchical Variational Autoencoder
Researchers have developed a causal hierarchical variational autoencoder (CHVAE) to generate realistic spine DXA images. This model is trained on UK Biobank data and conditioned on participant attributes and lumbar morphometry. The CHVAE demonstrates causal consistency by accurately synthesizing follow-up images after simulating age interventions, showing strong agreement with observed measurements. AI
IMPACT Demonstrates a new method for generating anatomically plausible medical images with causal consistency, potentially aiding in large-scale skeletal assessment and research.