A Temporal-Spectral Fused and Attention-Based Deep Model for Automatic Sleep Staging

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Automatic sleep staging is crucial for assessing sleep quality and diagnosing sleep disorders.
• Existing methods often overlook the dynamic relationships between sleep stages and primarily use time-domain data.

Purpose of the Study:

• To introduce TSA-Net, a novel deep neural network model for automatic sleep staging using single-channel electroencephalogram (EEG) signals.
• To enhance sleep staging accuracy by integrating temporal and spectral features and modeling stage transitions.

Main Methods:

• TSA-Net utilizes a two-stream feature extractor to fuse temporal and spectral EEG features.
• A multi-head self-attention mechanism in the feature context learning module captures dependencies between features.
• A conditional random field (CRF) module refines sleep stage classification by incorporating transition rules.

Main Results:

• TSA-Net achieved high accuracy rates of 86.64% and 82.21% on the Fpz-Cz channel in the Sleep-EDF-20 and Sleep-EDF-78 datasets, respectively.
• The model demonstrated superior performance compared to existing state-of-the-art methods in automatic sleep staging.

Conclusions:

• The proposed TSA-Net effectively optimizes sleep staging performance by leveraging fused temporal-spectral features and attention mechanisms.
• This deep learning approach offers a promising advancement for accurate and automated sleep quality assessment and clinical diagnosis.