AISleep: Automated and interpretable sleep staging from single-channel EEG data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Manual sleep staging is laborious and limits large-scale studies.
• Accurate sleep staging is crucial for diagnosing sleep disorders and understanding sleep physiology.

Purpose of the Study:

• To introduce AISleep, an automated, unsupervised sleep staging algorithm using a single electroencephalogram (EEG) channel.
• To evaluate AISleep's performance against state-of-the-art methods and assess its generalizability.

Main Methods:

• Developed AISleep, an unsupervised algorithm based on feature-weighted kernel density estimation (KDE).
• Validated AISleep on public benchmark datasets and clinical patient data.
• Compared AISleep with existing unsupervised and supervised sleep staging models.

Main Results:

• AISleep outperforms current unsupervised sleep staging algorithms in healthy young adults.
• The algorithm demonstrates superior generalizability compared to supervised models.
• Identified age-related decline in key EEG features impacting staging accuracy in older adults.

Conclusions:

• AISleep offers a robust, interpretable, and lightweight solution for automated sleep staging.
• The algorithm is suitable for integration into portable devices for scalable, home-based sleep monitoring.
• Findings highlight potential challenges in sleep staging accuracy for older populations.