Sleep Analysis Using Longitudinal Ear-EEG Recordings.

Area of Science:

• Neuroscience
• Sleep Medicine
• Biomedical Engineering

Background:

• Ear-electroencephalography (ear-EEG) is a less intrusive alternative to traditional scalp EEG for longitudinal sleep studies.
• Automated sleep analysis is crucial for the feasibility of large-scale, long-term sleep research.

Purpose of the Study:

• To develop and validate a systematic, automated procedure for sleep analysis using ear-EEG data.
• To focus on robust identification and exclusion of poor-quality sleep recordings without manual intervention.

Main Methods:

• Utilized the USleep sleep scoring model, leveraging its confidence score as a proxy for signal quality.
• Correlated the model's confidence score with Cohen's kappa to assess agreement with manual annotations.
• Evaluated the procedure on 574 ear-EEG sleep recordings from 24 chronic pain patients.

Main Results:

• The automated procedure achieved 91.9% accuracy in distinguishing recordings with high vs. low data quality (kappa > 0.6).
• Exclusion criteria did not disproportionately remove recordings with poor sleep metrics like low efficiency.
• Demonstrated the value of multi-night studies in visualizing inter-night sleep variability.

Conclusions:

• The proposed automated procedure effectively identifies and excludes poor-quality ear-EEG recordings for sleep analysis.
• This method enables more robust and reliable analysis of sleep patterns, particularly in patient populations.
• Supports the use of ear-EEG and automated analysis for efficient longitudinal sleep research.