Elimination of Muscle Artifacts from the Single Channel EEG Signal using Automatic Singular Spectrum Analysis

This paper introduces a novel approach for eliminating muscle artifacts from single-channel EEG signals by leveraging Automatic Singular Spectrum Analysis (ASSA) and statistical analysis. The method involves decomposing the contaminated EEG signals into reconstruction components (RCs) using ASSA, followed by an assessment of kurtosis measures for each RC to identify and isolate noisy components. Subsequently, the filtered EEG signal is reconstructed based on the linear combination of the remaining RCs. Through comprehensive evaluation using semi-synthetic and real EEG datasets sourced from public repositories like Biosource and Mendeley, the proposed approach demonstrates superior performance compared to existing methods. It achieves the lowest mean absolute error (MAE) in the power spectral density (PSD) domain while ensuring the preservation of essential clinical EEG features. By addressing a fundamental challenge in single-channel EEG processing, this method fills a crucial gap in biomedical signal analysis, particularly in ambulatory settings. Its simplicity and effectiveness make it a promising tool for real-world applications, challenging traditional paradigms in EEG signal denoising.