Sleep staging from single-channel EEG with multi-scale feature and contextual information

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Accurate sleep staging is crucial for diagnosing sleep disorders.
• Current methods often require complex, multi-sensor setups.
• Portable, high-accuracy devices can facilitate earlier diagnosis and treatment.

Purpose of the Study:

• To develop SleepStageNet, a novel single-channel electroencephalogram (EEG) sleep staging model.
• To leverage multi-scale convolutional neural networks (CNN) for EEG feature extraction.
• To utilize recurrent neural networks (RNN) and conditional random field (CRF) for contextual sleep stage inference.

Main Methods:

• Validation on two datasets: 20 healthy individuals (Fpz-Cz, Pz-Oz EEG) and 104 obstructive sleep apnea (OSA) patients (F4-M1 EEG).
• Sleep stages classified into four states: wake, REM, light sleep, and deep sleep.
• Performance evaluated using epoch-by-epoch accuracy and Cohen's kappa (ҡ) against polysomnography (PSG) scorers.

Main Results:

• SleepStageNet achieved high accuracy (0.88, 0.85) and kappa (0.82, 0.77) in healthy individuals.
• The model demonstrated strong performance in OSA patients with accuracy (0.80) and kappa (0.67).
• Significant improvements in accuracy and kappa were observed compared to existing methods.

Conclusions:

• SleepStageNet is a feasible tool for assessing sleep architecture in OSA patients using single-channel EEG.
• This advancement can enhance the utility of home sleep apnea testing.
• The model offers a promising approach for more accessible sleep disorder diagnosis.