Generalizable Deep Learning-Based Sleep Staging Approach for Ambulatory Textile Electrode Headband Recordings

Area of Science:

• Biomedical Engineering
• Neuroscience
• Sleep Medicine

Background:

• Automated sleep stage identification is crucial for clinical and research settings, particularly in home environments.
• Textile electrode headbands offer a user-friendly alternative to traditional sleep monitoring equipment.
• Previous research indicated textile headband signals resemble standard electrooculography (EOG).

Purpose of the Study:

• To develop and validate an automated sleep staging method using electroencephalographic (EEG) signals from a textile electrode headband.
• To assess the generalizability of a neural network model trained on polysomnography (PSG) data to ambulatory home recordings.
• To compare the performance of the textile headband method with standard EOG in home sleep studies.

Main Methods:

• A fully convolutional neural network (CNN) was trained and validated using a large clinical PSG dataset (n=876) with manually annotated sleep stages and standard EOG signals.
• The model's generalizability was tested on ambulatory home recordings from 10 healthy volunteers, comparing textile headband EEG with standard gel-based electrode EOG.
• Performance was evaluated using accuracy and Cohen's kappa (κ) for 5-stage sleep classification.

Main Results:

• The CNN model achieved 80% accuracy (κ=0.73) for 5-stage sleep classification using standard EOG in the clinical test set (n=88).
• The model generalized well to textile headband data, yielding 82% accuracy (κ=0.75).
• In home recordings, the model achieved 87% accuracy (κ=0.82) using standard EOG, demonstrating strong performance in an ambulatory setting.

Conclusions:

• The developed CNN model demonstrates significant potential for automated sleep staging in healthy individuals within a home environment.
• A reusable textile electrode headband can be effectively utilized for sleep monitoring, offering a convenient and user-friendly approach.
• This technology facilitates reliable, automated sleep analysis outside of traditional clinical settings, advancing sleep research and diagnostics.