Generalizable Sleep Staging via Multi-Level Domain Alignment

2023-12-13

AAAI Conference on Artificial Intelligence

AI Summary

Existing automated sleep staging models suffer from poor generalizability, exhibiting substantial performance degradation when evaluated across unseen datasets. To address this limitation, we propose a novel domain-generalizable paradigm for sleep staging—the first to explicitly formulate sleep staging as a domain generalization (DG) problem. Our method introduces a multi-level domain alignment mechanism that jointly learns epoch-level and sequence-level domain-invariant representations. It integrates unsupervised feature distribution matching with deep temporal modeling to enable end-to-end, label-efficient DG training without access to target-domain labels. Evaluated on five public benchmark datasets, our approach achieves state-of-the-art performance and demonstrates significantly improved generalization to entirely unseen datasets. This work establishes a transferable, robust technical framework for sleep staging, advancing the field toward clinically deployable, cross-population solutions.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Humanities & Computational Social Science
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Automatic sleep staging is essential for sleep assessment and disorder diagnosis. Most existing methods depend on one specific dataset and are limited to be generalized to other unseen datasets, for which the training data and testing data are from the same dataset. In this paper, we introduce domain generalization into automatic sleep staging and propose the task of generalizable sleep staging which aims to improve the model generalization ability to unseen datasets. Inspired by existing domain generalization methods, we adopt the feature alignment idea and propose a framework called SleepDG to solve it. Considering both of local salient features and sequential features are important for sleep staging, we propose a Multi-level Feature Alignment combining epoch-level and sequence-level feature alignment to learn domain-invariant feature representations. Specifically, we design an Epoch-level Feature Alignment to align the feature distribution of each single sleep epoch among different domains, and a Sequence-level Feature Alignment to minimize the discrepancy of sequential features among different domains. SleepDG is validated on five public datasets, achieving the state-of-the-art performance.

Problem

• Improving model generalization for unseen sleep datasets
• Aligning epoch-level features across different sleep domains
• Reducing sequential feature discrepancies among diverse sleep datasets

Innovation

• Multi-level feature alignment for domain generalization
• Epoch-level feature alignment for single sleep epochs
• Sequence-level feature alignment for sequential features