Ping He (Advisor), Nasser Kashou (Committee Member), Julie Skipper (Committee Member)
Master of Science in Biomedical Engineering (MSBME)
The biophysics of volume conduction that enable electrophysiological data acquisition also result in the mixing of data sources including possible, undesirable noise sources at the electrode interface. This work specifically focuses on improving the performance of the recursive least-squares (RLS) adaptive filtering method for removing eye movement artifact from the electroencephalogram. In biophysically-inspired simulated data, the RLS algorithm is verified to produce results that are inferior to extended infomax independent component analysis (ICA), the most widely used artifact correction approach in this problem space, due to its non-linear filter phase response and the presence of bidirectional contamination, or cross-talk, resultant of volume conduction in electroencephalographic data. The non-linear phase response of the RLS algorithm is mitigated by restricting its filter coefficients to form a linear phase, Type I finite impulse response filter. A reduced effect of cross-talk in RLS is achieved by filtering the reference noise input signal using a combination of non-local means weighting and Bayesian adaptive regression splines smoothing. When compared to extended infomax ICA, the modified RLS adaptive filtering approach meets or exceeds data source recovery accuracy while retaining highly desirable properties not afforded by blind source separation. These results support the use of a modified adaptive filtering approach for the near-ideal removal of eye artifact data from the electroencephalogram.
Department or Program
Department of Biomedical, Industrial & Human Factors Engineering
Year Degree Awarded
Copyright 2015, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.