Automatic Artifact Removal from EEG Signals using W-MOBICA

Abstract

Electroencephalography (EEG) signals are recurrently prone to be contaminated with several artifactual components instinctive from the eye movements, cardiac, muscle contractions and bad electrode. These artifacts yield an increase in amplitude components and alter the original EEG recordings, misleading the neurologists in diagnosing neural disorders. This paper presents an automatic artifact identification and removal strategy without consuming any supplementary reference channel. In this work, an ICA-Wavelet decomposition approach in conjunction with a threshold based peak detection and removal strategy is constructed. ICA variant approach used in our paper is Online Recursive Independent Component Analysis (ORICA). Usually, in ORICA, Recursive Least Square (RLS) formulation is used, but RLS suffers from high computational complexity. In our work, we replaced RLS by Bi-Conjugate Gradient (BCG) algorithm and the design parameter is modified using the RLS normal equation which accelerates the convergence rate and reduces the complexity of the proposed model. Our Wavelet enhanced Modified Online Bi-Conjugate Gradient based Independent Component Analysis (W-MOBICA) removes the artifactual component parting the intact EEG source component. Empirical results on CHB-MIT and SRM databases of 52 EEG recordings outperforms the state-of-the-art approaches maintaining least MAE, RMSE and high SAR, MI, CC. The proposed strategy can be a promising solution for artifact removal in the clinical use of EEG signals and in the Brain Computer Interface (BCI) applications.