Artifact Removal from EEG signals using Regenerative Multi-Dimensional Singular Value Decomposition and Independent Component Analysis

The EEG signals are regularly blended with sources like Electrooculogram, Electromyogram and few other ar-tifacts caused by physical or signal interferences. The presence of artifacts induces inaccuracy in the examination of the signals acquired. Independent Component Analysis has been predominantly utilized towards these dis-crepancies by isolating the artifacts from the EEG signals. Direct utilization of ICA isn't conceivable with the frameworks that are outfitted with single or few EEG channels. Distinctly using ICA to eliminate artifacts on a single channel is harder. Therefore, we combine ICA with a proposed decomposition method called Regenerative Multi-Dimensional Singular Value Decomposition (RMD-SVD) which maps the acquired signals into multivariate data after which ICA is applied on it. In our proposed scheme, the pattern of a source signal is mimicked with frequency, phase and amplitude value of the input signal using EEG sigmoid function. Both the input signal and the constructed regenerative reference signals are decomposed and the most significant singular values can be observed with the help of ICA which are the values of the pure input signal. Performance measures such as SNR, PSNR, MSE etc., in our proposed systems are analyzed under different filters and it is noticed that our proposed method of RMD-SVD indicates increased noise omitting efficiency.

Automated summary

This study presents a method that combines Regenerative Multi-Dimensional Singular Value Decomposition (RMD-SVD) with Independent Component Analysis (ICA) to eliminate artifacts in EEG signals. By mapping the signals into multivariate data and applying ICA, accurate separation of artifacts can be achieved. Experimental results demonstrate that the proposed RMD-SVD method significantly improves noise removal efficiency.