Segmentation of EEG during sleep using time-varying autoregressive modeling

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Sleep electroencephalogram (EEG) analysis is crucial for understanding sleep stages and detecting abnormalities.
• Traditional autoregressive (AR) models assume signal stationarity, which is often violated in biological signals like EEG.
• Time-varying modeling offers a more flexible approach to capture dynamic changes in EEG characteristics.

Purpose of the Study:

• To apply time-varying AR (TV-AR) modeling for parameter estimation and segmentation of sleep EEG signals.
• To evaluate different basis functions for their efficiency in approximating EEG parameter changes.
• To develop a segmentation algorithm based on estimated TV-AR model parameters.

Main Methods:

• Analysis of various basis functions for approximating EEG signal parameter variations.
• Application of the TV-AR model to simulated sleep EEG segments under diverse conditions.
• Examination of estimation behavior concerning parameter variations, boundary locations, and basis function orders.
• Development of an "Identification function" for segmentation.

Main Results:

• Discrete Cosine Transform (DCT) and Walsh functions demonstrated the highest efficiency in estimating TV-AR model parameters for sleep EEG.
• The performance of parameter estimation was evaluated under varying EEG parameters and segment boundary locations.
• The study identified optimal basis functions for accurate sleep EEG signal characterization.

Conclusions:

• TV-AR modeling is a viable technique for analyzing dynamic changes in sleep EEG.
• DCT and Walsh functions are recommended for efficient parameter estimation in sleep EEG analysis.
• The proposed segmentation algorithm based on the "Identification function" shows promise for automated sleep EEG analysis.