Sleep-spindle identification on EEG signals from polysomnographie recordings using correntropy

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Sleep spindles (SSs) are key electroencephalographic (EEG) waveforms in N2 and N3 sleep stages.
• Accurate SS detection is crucial for sleep analysis, but high false positive rates pose a significant challenge.

Purpose of the Study:

• To develop a novel periodogram using correntropy for enhanced sleep spindle detection and characterization.
• To improve the accuracy of SS detection by reducing false positives.

Main Methods:

• A new periodogram was developed based on correntropy, a generalized correlation measure from information theoretic learning.
• Non-negative matrix factorization decomposition of correntropy was employed to create the new periodogram.
• The proposed method's performance was evaluated against conventional methods.

Main Results:

• The correntropy-based periodogram demonstrated improved resolution compared to the conventional power spectrum density.
• Preliminary results indicated a high sensitivity rate of 0.868.
• The method achieved a low false positive rate of 0.121.

Conclusions:

• The proposed correntropy-based periodogram offers a promising approach for accurate sleep spindle detection.
• This method enhances SS characterization and significantly reduces false positives, aiding sleep research.