Omni-Sleep Model Improves Sleep Analysis with Physiological Hierarchy.

A new sleep foundation model, Omni-Sleep, has been developed to enhance the analysis of sleep physiology by incorporating the inherent hierarchical structure of the central nervous system (CNS) and autonomic nervous system (ANS). Unlike previous models that often fuse biosignals without considering their physiological relationships, Omni-Sleep uses this structure as a prior for learning. The model employs three key objectives: ensuring consistency within each system (CNS and ANS), synchronizing trajectories between the systems to capture brain-body dynamics, and using masked temporal modeling to understand long-term sleep patterns. This approach allows Omni-Sleep to learn structured representations from multimodal polysomnography data, including EEG, EOG, EMG, ECG, and respiration. Evaluated on over 100,000 hours of multi-center data, Omni-Sleep demonstrates superior performance in tasks like sleep staging and multi-disease classification. It shows improved label efficiency, better generalization across different datasets, and increased robustness even when some modalities are missing, highlighting the benefits of its physiologically informed design.