A novel, fast and efficient single-sensor automatic sleep-stage classification based on complementary cross-frequency...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Manual sleep stage scoring is time-consuming and requires extensive inter-rater reliability checks.
• There is a growing need for automated methods in sleep analysis due to the limitations of manual scoring.
• Automatic Sleep Stage Classification (ASSC) using signal processing and machine learning is of significant interest.

Purpose of the Study:

• To present a novel single-EEG-sensor technique for Automatic Sleep Stage Classification (ASSC).
• To evaluate the performance of the proposed ASSC method using cross-frequency coupling (CFC) analysis.
• To demonstrate the applicability of the ASSC method for longitudinal sleep monitoring.

Main Methods:

• Developed a single-EEG-sensor ASSC technique utilizing dynamic reconfiguration of cross-frequency coupling (CFC).
• Estimated CFC between predefined frequency pairs over 5-second epoch lengths.
• Validated the methodology on the PhysioNet Sleep European Data Format (EDF) Database and a second sleep dataset.

Main Results:

• Achieved high classification performance with sensitivity (96.2%), specificity (94.2%), and accuracy (94.4%) across 20 folds.
• Obtained a high mean F1 score of 92% (range 90-94%) using a multi-class Naive Bayes classifier.
• Demonstrated superior classification accuracy compared to previous studies on multiple datasets.

Conclusions:

• The proposed single-sensor ASSC method offers high accuracy and outperforms existing approaches.
• The technique is suitable for longitudinal sleep monitoring in both real-world and laboratory settings.
• The methodology facilitates the use of wearable EEG devices for continuous sleep analysis.