Sleep-wake stages classification based on single channel ECG signals by using a dynamic connection convolutional...

Area of Science:

• Sleep Medicine
• Biomedical Engineering
• Signal Processing

Background:

• Accurate sleep-wake stage identification is vital for assessing sleep quality.
• Current methods often rely on electroencephalogram (EEG) signals, which are inconvenient and susceptible to noise.
• Electrocardiogram (ECG) signals offer a simpler, more convenient alternative for sleep monitoring.

Purpose of the Study:

• To develop a simple and effective sleep-wake stage classification model using ECG signals.
• To enable sleep monitoring through wearable devices.
• To overcome the limitations of EEG-based methods.

Main Methods:

• Extraction of multi-scale ECG signal features using convolutional kernels of varying sizes.
• Proposal of a novel dynamic connection convolutional neural network (DCCNN) for classification.
• DCCNN utilizes layer feature map 'goodness' to form optimal residual modules.

Main Results:

• The proposed DCCNN model achieved a peak accuracy of 92.21% on the MIT-BIH Polysomnographic Database.
• Performance is comparable and often superior to traditional EEG-based sleep-wake classification methods.
• The approach demonstrates significant effectiveness compared to existing state-of-the-art techniques.

Conclusions:

• This research presents a novel, effective, and convenient approach for sleep monitoring using ECG signals.
• The DCCNN model offers a promising alternative for wearable sleep tracking applications.
• The findings support the potential of ECG-based analysis for sleep medicine.