SleepMaMi: A Universal Sleep Foundation Model for Integrating Macro- and Micro-structures

SleepMaMi is introduced as a universal Sleep Foundation Model designed to integrate both macro- and micro-structural features of sleep, utilizing a hierarchical dual-encoder architecture. The Macro-Encoder captures full-night temporal dependencies through Demographic-Guided Contrastive Learning, while the Micro-Encoder employs a hybrid Masked Autoencoder and multi-modal contrastive objectives, trained on over 20,000 PSG recordings. This model demonstrates improved generalizability and label-efficient adaptation for clinical sleep analysis, marking a significant advancement in the application of foundation models in sleep medicine.