Automatic Sleep Stage Classification Using Temporal Convolutional Neural Network and New Data Augmentation Technique...

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence

Background:

• Automatic sleep stage classification is crucial for monitoring sleep disorders.
• Single-channel electroencephalography (EEG) offers a promising avenue for non-invasive sleep monitoring.
• Current methods often rely on multi-channel EEG or manual analysis, limiting accessibility.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for automatic sleep stage classification using single-channel EEG signals.
• To improve the accuracy and efficiency of sleep disorder monitoring.
• To enable the development of wearable sleep monitoring devices.

Main Methods:

• A deep learning architecture combining Convolutional Neural Networks (CNNs) for feature extraction and Temporal Convolutional Networks (TCNs) for temporal feature analysis was employed.
• A Conditional Random Field (CRF) layer was integrated to refine classification performance.
• A novel data augmentation technique was utilized to enhance CNN training.

Main Results:

• The proposed model demonstrated superior performance on two public sleep datasets (Sleep-EDF-2013 and Sleep-EDF-2018) using single-channel EEG data.
• Achieved high total accuracy (85.39% on EDF-2013, 82.46% on EDF-2018) and kappa scores (0.80 and 0.76, respectively).
• Outperformed existing state-of-the-art methods in sleep stage classification.

Conclusions:

• The developed deep learning framework shows significant potential for accurate automatic sleep stage classification from single-channel EEG.
• The model's ability to learn patterns directly from data minimizes engineering bias compared to traditional methods.
• This research facilitates the development of practical, wearable sleep monitoring systems for broader clinical and personal use.