Reproduce “Multimodal Cardiovascular Risk Profiling Using Self-Supervised Learning of Polysomnography” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhengxiao He, Huayu Li, Geng Yuan, William D.S. Killgore, Stuart F. Quan, Chen X. Chen, Ao Li

A self-supervised deep learning framework developed at the University of Arizona extracts interpretable physiological patterns from raw multimodal polysomnography (PSG) data to profile cardiovascular risk. This approach significantly improved prediction of prevalent and incident cardiovascular outcomes and mortality, achieving AUCs up to 0.965 for atrial fibrillation when combined with traditional risk scores.

Key insights

  • Self-supervised learning on raw multimodal PSG effectively captures latent disease-relevant physiological patterns, removing the dependency on laborious manual annotations.
  • ECG-derived projection scores are exceptionally robust predictors for direct cardiac conditions, while EEG-derived scores show associations with CVD mortality and hypertension, suggesting novel digital biomarkers.
  • The integration of PSG-derived projection scores with established clinical risk factors, particularly the Framingham Risk Score, consistently and substantially improves the prediction of a wide range of cardiovascular outcomes.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.