Two-Stage Atomic Decomposition of Multichannel EEG and the Previously Undetectable Sleep Spindles

Area of Science:

• Neuroscience
• Signal Processing
• Sleep Medicine

Background:

• Sleep spindles are crucial for memory consolidation.
• Current methods for detecting sleep spindles in electroencephalograms (EEGs) have limitations in sensitivity.
• Identifying all relevant EEG structures is essential for accurate sleep staging and analysis.

Purpose of the Study:

• To develop and validate a novel two-step procedure for atomic decomposition of multichannel EEGs.
• To improve the detection rate of sleep spindles compared to human scorers and existing algorithms.
• To characterize the spatial and temporal properties of detected sleep spindles, including frequency gradients.

Main Methods:

• A two-step procedure combining multivariate matching pursuit and dipolar inverse solution for EEG decomposition.
• Application of the method to 147 polysomnographic recordings from the Montreal Archive of Sleep Studies.
• Comparison of the proposed method's sleep spindle detection performance against human scorers and two state-of-the-art algorithms.

Main Results:

• The proposed method detected approximately three times more sleep spindles than existing methods and human scorers.
• Previously undetectable sleep spindles were identified, sharing properties with known spindles but obscured by noise.
• Automatic parametrization of detected EEG structures allowed for the estimation of spatial gradients in sleep spindle frequencies.

Conclusions:

• The novel atomic decomposition method significantly enhances the detection of sleep spindles in multichannel EEGs.
• The findings suggest that many sleep spindles are missed due to local signal-to-noise ratios, not inherent differences.
• This method provides new insights into the spatial distribution of sleep spindle frequencies, revealing sagittal plane gradients.