BTCRSleep: a boundary temporal context refinement-based fully convolutional network for sleep staging with
Caihong Zhao1, Jinbao Li2, Yahong Guo3
1School of Electronic and Engineer, School of Computer Science and Technology, Heilongjiang University, Harbin, 150080, People's Republic of China.
Physiological Measurement
|June 2, 2023
Summary
This study introduces BTCRSleep, a deep learning model that improves sleep staging by capturing boundary context in electroencephalogram (EEG) data. The model enhances accuracy in identifying sleep stages, especially during transitions.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Biomedical Engineering
Background:
- Single-channel electroencephalogram (EEG) sleep staging commonly uses deep learning, combining convolutional neural networks (CNNs) and recurrent neural networks (RNNs).
- CNNs may lose crucial boundary context information when sleep stage features span multiple epochs, impacting sleep staging accuracy.
Purpose of the Study:
- To improve sleep staging performance by capturing boundary temporal context during sleep stage transitions.
- To address the loss of contextual information in traditional deep learning approaches for EEG analysis.
Main Methods:
- Proposed a fully convolutional network with boundary temporal context refinement (BTCRSleep).
- Developed a boundary temporal context refinement module to enhance multi-scale temporal dependencies between epochs.
- Implemented a class-aware data augmentation method to improve learning of minority class boundary contexts.
Main Results:
- BTCRSleep achieved superior total accuracy and kappa scores compared to state-of-the-art methods on four public datasets (SEDF, SEDFX, SHHS, CAP).
- Achieved average accuracies of 84.9% (SEDF), 82.9% (SEDFX), 85.2% (SHHS), and 76.9% (CAP) under subject-independent cross-validation.
- Demonstrated that boundary temporal context significantly improves the capture of temporal dependencies across epochs.
Conclusions:
- The proposed BTCRSleep model effectively captures boundary temporal context for improved sleep staging.
- The findings highlight the importance of temporal context in sleep stage transition analysis using EEG data.
