An Efficient Sleep Scoring Method using Visibility Graph and Temporal Features of Single-Channel EEG.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Manual sleep scoring is time-consuming and requires expert analysis.
• Accurate sleep stage classification is vital for diagnosing sleep disorders and neurological conditions.

Purpose of the Study:

• To develop an automated method for sleep stage identification using electroencephalography (EEG) data.
• To fuse graph-based and temporal features for enhanced classification accuracy.

Main Methods:

• EEG epochs were converted into visibility graphs to extract graph-based features (mean degree, degree distribution).
• Temporal features including autoregressive parameters, fractal dimension, entropy, and Hjorth's parameters were calculated.
• An ensemble classifier (random undersampling with boosting) was employed using features from a single EEG channel (Pz-Oz).

Main Results:

• The model achieved high accuracies: 91.0% for 6-state and 97.3% for 2-state sleep classification.
• Kappa coefficients of 0.82 (6-state) and 0.94 (2-state) were obtained, outperforming existing methods.
• Performance was validated using 10-fold cross-validation and a 50% holdout approach.

Conclusions:

• The proposed fusion of graph and temporal EEG features offers a robust and accurate method for automatic sleep stage identification.
• This automated approach can significantly reduce the workload for sleep experts and aid in the diagnosis of sleep and consciousness disorders.