Global EEG segmentation using singular value decomposition.

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Analyzing multichannel electroencephalography (EEG) data requires accurate segmentation.
• Identifying temporal blocks with fixed spatial distributions of neuronal generators is crucial for understanding brain activity.

Purpose of the Study:

• To develop and validate a method for segmenting EEG data into stable temporal blocks.
• To accurately detect segment boundaries based on changes in neuronal generator spatial distributions.

Main Methods:

• Singular value decomposition (SVD) was employed for data segmentation.
• Segment boundaries were identified by statistically comparing residual errors from reference and sliding windows projected onto a feature subspace.
• The Kolmogorov-Smirnov (K-S) test was used for statistical testing, with aggregated decisions to enhance reliability.

Main Results:

• The proposed algorithm successfully detected segment boundaries in multichannel EEG data.
• Simulations confirmed the algorithm's effectiveness under various conditions.
• The method reliably segments data into blocks with fixed spatial distributions of neuronal generators.

Conclusions:

• The SVD-based method provides a reliable approach for segmenting EEG data.
• Accurate segmentation enhances the analysis of brain activity by isolating periods of stable neuronal configurations.
• This technique offers a valuable tool for neuroscience research and clinical applications.