An automatic single-channel EEG-based sleep stage scoring method based on hidden Markov Model.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Accurate sleep stage scoring is critical for diagnosing sleep disorders.
• Manual visual scoring is time-consuming and prone to human error.
• Developing automated, accurate sleep analysis methods is essential.

Purpose of the Study:

• To introduce an efficient approach for improving sleep stage scoring and classification accuracy.
• To enhance the reliability of sleep analysis using automated methods.
• To provide a robust solution for real-time sleep disorder diagnosis.

Main Methods:

• Optimal feature selection from sleep EEG epochs using relevance-redundancy analysis.
• Classification of EEG segments with a random forest classifier.
• Application of a Hidden Markov Model (HMM) to refine sleep stage transitions and reduce false positives.

Main Results:

• Achieved high accuracies (79.4-87.4% for six-stage, 77.6-80.4% for five-stage) with Kappa values (0.7-0.85) using leave-one-out validation.
• Demonstrated robust performance across multiple public EEG datasets (R&K and AASM guidelines).
• Observed up to an 8% reduction in accuracy with cross-dataset validation compared to subject cross-validation.

Conclusions:

• The proposed single-channel method offers robust and reliable sleep stage scoring.
• Achieved high accuracy with relatively low complexity, suitable for real-time applications.
• Outperformed existing methods in multi-stage sleep classification accuracy.