A Gait Foundation Model Predicts Multi-System Health Phenotypes from 3D Skeletal Motion

Summary

A new arXiv paper (2603.25283) by Adam Gabet, Sarah Kohn, Guy Lutsker, Shira Gelman, Anastasia Godneva and colleagues presents a gait foundation model trained on 3D skeletal motion data from 3,414 deeply phenotyped adults. Treating gait not merely as a symptom of specific pathologies but as a systemic biomarker, the learned embeddings outperform hand-engineered features, predicting age (Pearson r=0.69), BMI (r=0.90), and visceral adipose tissue area (r=0.82). The embeddings significantly predict 1,980 of 3,210 phenotype targets. Anatomical ablation studies reveal that leg movements primarily encode metabolic and frailty phenotypes, while the torso encodes sleep and lifestyle phenotypes. The authors argue these findings establish gait as an independent multi-system biosignal, positioning gait analysis via wearable or camera-based motion capture as a broad health-screening tool. This forum post summarizes the paper's abstract and links to the arXiv preprint.