Quantitative evaluation of techniques for ocular artefact filtering of EEG waveforms

Abstract

The electrical dipoles of eyes change by eye movements and blinks, producing a signal known as an electrooculogram (EOG). A fraction of EOGs contaminate the electrical activity of the brain (electroencephalogram, EEG). Ocular artefact (OA) is a collective term used to represent EEG contaminating potentials caused by eye movements and blinks. A procedure for quantifying the effectiveness of an algorithm for removing OA from the EEG was devised. This enabled the similarity between the EEG waveforms before contamination by OA and the contaminated EEG waveforms following their processing by an OA removal method to be measured. Four methods for OA removal were included in the study: extended independent component analysis (ICA), joint approximation diagonalisation of eigenmatrices (JADE), principal component analysis (PCA) and EOG subtraction. The operation of JADE and ICA is subject to amplitude scaling and channel permutation. Procedures were incorporated to estimate the amplitude of the recovered EEG waveforms and to allocate them to the correct channels. It was demonstrated that the signal separation techniques of JADE and extended ICA were more effective than EOG subtraction and PCA for removing OA from the EEG. EOG subtraction was shown to cause attenuation of the recovered EEG waveforms. The effect of additive Gaussian noise on the performance of the four OA removal methods was also investigated. This indicated that the performance of the methods was unaffected by an additive Gaussian noise source, as long as the signal-to-noise ratio remained above 50.