SVD-Based Feature Extraction Technique for The Improvement of Effective Connectivity Detection

Abstract

Electroencephalogram (EEG) plays an essential part in identifying brain function and behaviors for different mental states. Nevertheless, the captured electrical activity is always found to be contaminated with various artifacts that negatively influence the accuracy of EEG analysis. Therefore, it is crucial to build a model to constructively identify and extract clean EEG recordings during the investigation of the dynamical brain networks. To improve the estimation of effective connectivity (EC) and EEG signal denoising, an EEG decomposition method based on the singular value decomposition (SVD) analysis was proposed. The main purpose of the decomposition is to create a method to estimate a signal that represents most of the principal components of the information contained in each brain region before calculating the partial directed coherence (PDC). SVD-based technique and PDC were used to quantify the causal influence of default mode network (DMN) regions on each other and track the changes in brain connectivity. Results of statistical analysis on the effective connectivity using the SVD-PDC algorithm have shown to better reflect the flow of causal information than the independent component analysis (ICA)-PDC. The hybrid algorithm (SVD-PDC) is proposed in this work as an alternative robust adaptive feature extraction method for EEG signals to improve the detection of brain effective connectivity.