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

TL;DR

SleepMaMi is a comprehensive sleep foundation model that integrates macro- and micro-structure analysis of sleep using hierarchical encoders, trained on extensive PSG data, to improve clinical sleep analysis.

Contribution

The paper introduces SleepMaMi, a novel hierarchical dual-encoder framework that captures both global sleep architecture and local biosignal features, trained on a large-scale PSG dataset.

Findings

Outperforms existing models on multiple sleep analysis tasks.

Demonstrates superior generalizability and label efficiency.

Effectively models full-night sleep macro-structures and micro-structures.

Abstract

While the shift toward unified foundation models has revolutionized many deep learning domains, sleep medicine remains largely restricted to task-specific models that focus on localized micro-structure features. These approaches often neglect the rich, multi-modal context of Polysomnography (PSG) and fail to capture the global macro-structure of a full night's sleep. To address this, we introduce SleepMaMi , a Sleep Foundation Model engineered to master both hour-long sleep architectures and fine-grained signal morphologies. Our framework utilizes a hierarchical dual-encoder design: a Macro-Encoder to model full-night temporal dependencies and a Micro-Encoder to capture short-term characteristics from biosignals. Macro-Encoder is trained via Demographic-Guided Contrastive Learning, which aligns overnight sleep patterns with objective subject metadata, such as age, sex and BMI to refine global representations. Micro-Encoder is optimized via a hybrid Masked Autoencoder (MAE) and multi-modal contrastive objective. Pre-trained on a massive corpus of $>$20,000 PSG recordings (158K hours),SleepMaMi outperforms existing foundation models across a diverse suite of downstream tasks, demonstrating superior generalizability and label-efficient adaptation for clinical sleep analysis.