Sleep Staging Algorithm Incorporating Multi-Threhold Neighborhood Polar Pattern Statistics.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Sleep quality is critical for overall health and is evaluated through sleep staging.
• Accurate sleep staging relies on analyzing electroencephalogram (EEG) signals.
• Existing methods require improvement for enhanced precision and reliability.

Purpose of the Study:

• To propose a novel sleep staging method based on statistical mode of multi-threshold neighborhood extreme (SMNE).
• To enhance EEG signal preprocessing using discrete wavelet transform (DWT) and data augmentation.
• To evaluate the performance of the proposed SMNE method on benchmark sleep datasets.

Main Methods:

• EEG signals were preprocessed using discrete wavelet transform (DWT) and a data enhancement algorithm.
• Signal-to-noise ratio (SNR) assessment and signal overlap analysis were employed for quality improvement.
• Extremes of EEG signals were classified into 5 states, with patterns extracted using multi-thresholding and statistical encoding.
• Grey wolf optimization (GWO) was used for threshold determination, followed by Random Forest (RF) classification.

Main Results:

• The SMNE method achieved high accuracy on multiple datasets: 94.6% (SleepEDFx), 96.3% (SleepEDF-20), and 88.5% (ISRUC-Sleep).
• Excellent performance was also demonstrated by kappa coefficients (0.92, 0.94, 0.83) and F1-scores (89.3%, 94.2%, 86.5%).
• The proposed method effectively extracts and classifies sleep-related patterns from EEG data.

Conclusions:

• The SMNE-based sleep staging method offers a robust and accurate approach for sleep quality assessment.
• The integration of DWT, SNR analysis, and GWO optimization significantly improves EEG signal analysis.
• This novel method holds promise for advancing sleep research and clinical diagnostics.