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

2026-02-07

AI Summary

This work proposes SleepMaMi, a unified foundation model for sleep analysis that addresses the limitations of existing approaches, which often focus narrowly on task-specific microstructural features and fail to capture the macroscopic temporal structure and contextual information across whole-night, multimodal polysomnography (PSG) signals. SleepMaMi introduces a novel dual-encoder architecture comprising a macro-encoder to model the global sleep architecture over an entire night and a micro-encoder to learn fine-grained morphological patterns in biosignals. The model is pretrained on over 20,000 PSG recordings (158,000 hours) using demographic-informed contrastive learning, hybrid masked autoencoding, and multimodal contrastive learning. Evaluated across diverse downstream tasks, SleepMaMi demonstrates superior performance, remarkable generalization capability, and high label efficiency, advancing the automation of clinical sleep analysis.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: (Computational) Cognitive Architectures
• Computer Vision: Multi-modal Vision

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

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.

Problem

• sleep foundation model
• macro-structure
• micro-structure
• polysomnography
• multi-modal context

Innovation

• Sleep Foundation Model
• Macro-Micro Integration
• Hierarchical Dual-Encoder
• Demographic-Guided Contrastive Learning
• Masked Autoencoder