Self-evaluated automatic classifier as a decision-support tool for sleep/wake staging.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Polysomnography (PSG) signals often contain artifacts that hinder classification accuracy.
• Existing automatic systems may lack robustness in handling artifacts and providing reliable confidence measures.

Purpose of the Study:

• To develop and validate a two-stage automatic sleep/wake classifier robust to signal artifacts.
• To incorporate a confidence index into the classification output to enhance user trust and guide expert review.
• To improve the discrimination between specific sleep stages, such as NREM I and REM sleep.

Main Methods:

• A two-stage classification approach was implemented.
• Stage 1: Artifact detection and selection of artifact-free epochs from polysomnographic signals (EEG, EOG, EMG).
• Stage 2: Classification of artifact-free epochs using one of four selected classifiers, with a confidence index generated based on classifier and assigned class.

Main Results:

• The system achieved an overall accuracy of 85.5% on a large database of 46 night recordings.
• Demonstrated improved ability in discerning the NREM I stage from REM sleep.
• Only 7% of the data was classified with low confidence, indicating high reliability and efficiency as a decision-support tool.

Conclusions:

• The proposed two-stage automatic classifier effectively handles artifacts in polysomnographic data.
• The integrated confidence index enhances the system's utility as a decision-support tool for sleep analysis.
• This approach offers a reliable and efficient method for automatic sleep staging, reducing the burden on human experts.