Communications in Information and Systems

Volume 3 (2003)

Number 1

EEG ocular artifact removal through ARMAX model system identification using extended least squares

Pages: 19 – 40

DOI: http://dx.doi.org/10.4310/CIS.2003.v3.n1.a2

Authors

Shane M. Haas (Laboratory for Information and Decision Systems, and Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, Mass., U.S.A.)

Mark G. Frei (Flint Hills Scientific, L.L.C., Lawrence, Kansas, U.S.A.)

Ivan Osorio (Comprehensive Epilepsy Center, Kansas City, Kansas, U.S.A.; Flint Hills Scientific, L.L.C., Lawrence, Kansas, U.S.A.)

Bozenna Pasik-Duncan (Department of Mathematics, University of Kansas, Lawrence, Ks., U.S.A.)

Jeff Radel (University of Kansas Medical Center, Kansas City, Kansas, U.S.A.)

Abstract

The removal of ocular artifact from scalp electroencephalograms (EEGs) is of considerable importance for both the automated and visual analysis of underlying brainwave activity. Traditionally, subtraction techniques use linear regression to estimate the influence of eye movements on the electrodes of interest. These methods are based on the assumption that the underlying brainwave activity is uncorrelated when, in general, it is not. Furthermore, regression methods assume that the ocular artifact propagation is frequency independent, i.e. all waveforms of the ocular artifact propagate similarly. In this paper, we examine relaxing these assumptions by using a more general autoregressive (AR) moving average (MA) exogenous (X) model and the extended least squares (ELS) algorithm to remove ocular artifact. We demonstrate that in some cases this general ARMAX model can decrease ocular artifact not removable by standard regression techniques. We also show that the incorporation of a forgetting factor to exponentially weight past data can improve ocular artifact removal even for the traditional subtraction method.

Keywords

ocular artifact removal, system identification, extended least squares, adaptive noise cancellation, epilepsy, electroencephalogram signal processing, weighted least squares

Full Text (PDF format)