Removing muscle artifacts from EEG data of people with cognitive impairment using high order statistic methods

TL;DR

This paper introduces a novel method combining second and high order statistical information to effectively remove muscle artifacts from EEG data, enhancing the analysis of brain activity in cognitively impaired individuals.

Contribution

A new approach using joint Blind Source Separation with second and high order statistics, improved by SCHUR decomposition for faster processing, specifically for EEG artifact removal.

Findings

Effective removal of muscle artifacts demonstrated on simulated EEG data.

Improved signal quality in real EEG data from cognitively impaired subjects.

Method outperforms traditional techniques in preserving brain signals.

Abstract

Objective: Often, people with Subjective Cognitive Impairment (SCI), Mild Cognitive Impairment (MCI) and dementia are underwent to Electroencephalography (EEG) in order to evaluate through biological indexes the functional connectivity between brain regions and activation areas during cognitive performance. EEG recordings are frequently contaminated by muscle artifacts, which obscure and complicate their interpretation. These muscle artifacts are particularly difficult to be removed from the EEG in order the latter to be used for further clinical evaluation. In this paper, we proposed a new approach in removing muscle artifacts from EEG data using a method that combines second and high order statistical information. Subjects and Methods: In the proposed system the muscle artifacts of the EEG signal are removed by using the Independent Vector Analysis (IVA). The latter was formulated as a general joint Blind Source Separation (BSS) method that uses both second-order and higher order statistical information and thus takes advantage of both Independent Component Analysis (ICA) and Canonical Correlation Analysis (CCA). Diagonalization methods for IVA in the proposed system were reworked based on SCHUR decomposition offering a faster second order blind identification algorithm that can be used on time demanding applications. Results: The proposed method is evaluated in both simulated and real EEG data. To quantitatively examine the performance of the new method, two objective measures were adopted. The first measure is the Root Mean Square Error (RMSE) while the second is the Signal-to-noise-ratio (SNR). Conclusion: The proposed method overcomes with the need of removing muscle artifacts on both realistic simulated EEG data and brain activity from people with cognitive impairment.