TRA: Trajectory retrieval augmented model for automatic sleep staging

doi:10.1016/j.bspc.2025.108975

TRA: Trajectory retrieval augmented model for automatic sleep staging

2025-11-05

0

PRE

AI

DOI：10.1016/j.bspc.2025.108975

Original

Original request for help

Save

Abstract

En 中文

Automatic sleep staging plays a crucial role in sleep quality assessment and the diagnosis of sleep disorders. Although existing deep learning-based approaches have advanced this task to some extent, their further application and research are still severely constrained by the scarcity of clinically annotated data. To address this issue, we propose a Trajectory Retrieval Augmented (TRA) framework. Specifically, a trajectory retrieval mechanism is introduced to alleviate the data scarcity problem by retrieving highly similar, category-consistent sequences. Furthermore, a Siamese multi-scale feature extraction module is designed to encode more discriminative category-consistent representations, thereby substantially enhancing model performance. Comparative experiments against five baseline methods demonstrate that the proposed approach achieves improvements of 3.84% in the macro-F1 score, 2.75% in accuracy and 3.21% in Cohen’s Kappa. These results confirm that our method effectively mitigates the performance bottleneck caused by data scarcity, thereby enabling more accurate sleep staging for improved sleep quality assessment. The source code is publicly available at https://github.com/xusheng1234567/TRA .

TRA: Trajectory retrieval augmented model for automatic sleep staging

Abstract

Journal

Organization