Model-Based Electroencephalogram Instantaneous Frequency Tracking: Application in Automated Sleep-Wake Stage...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate sleep stage classification is vital for diagnosing sleep disorders and understanding sleep's health impacts.
• Analyzing electroencephalogram (EEG) data, the gold standard for sleep monitoring, is time-consuming and resource-intensive.
• Automated EEG analysis offers a cost-effective solution for efficient sleep data interpretation.

Purpose of the Study:

• To develop a novel EEG feature extraction pipeline for accurate wake and sleep stage classification.
• To implement a noise-robust Kalman filtering approach for tracking time-varying autoregressive models in EEG data.
• To create a two-step classifier for detailed sleep staging, differentiating wake, REM, and various non-REM stages.

Main Methods:

• Utilized a model-based Kalman filtering (KF) approach to track time-varying autoregressive (TVAR) models in EEG signals.
• Extracted key EEG features, including instantaneous frequency and power.
• Employed a two-step classification strategy: first, categorizing into wake, REM, and non-REM; second, sub-classifying non-REM into N1, N2, and N3 stages.

Main Results:

• Evaluated on the Sleep-EDFx dataset (153 recordings, 78 subjects).
• Achieved over 77% overall accuracy with Light Gradient Boosting Machine (LGBM) and Extreme Gradient Boosting (XGBoost) classifiers.
• Obtained a macro-averaged F1 score of 0.69 and Cohen's kappa of 0.68, demonstrating robust performance.

Conclusions:

• The proposed novel EEG feature extraction pipeline effectively classifies wake and sleep stages.
• The Kalman filtering and TVAR model approach provides a compact and interpretable feature set for sleep analysis.
• This method offers a promising solution for automated, efficient, and accurate sleep staging, aiding in sleep disorder diagnosis and research.