Higher Order Statistics-Based Radial Basis Function Network for Evoked Potentials

• Published in: IEEE transactions on biomedical engineering Vol. 56; no. 1; pp. 93 - 100
• Main Authors: Lin $^$, Bor-Shyh, Lin, Bor-Shing, Chong, Fok-Ching, Lai, Feipei
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2009; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Brain - physiology; Brain modeling; Computer Simulation; Data Interpretation, Statistical; Data mining; Electroencephalography; Electroencephalography - methods; Evoked potentials; Evoked Potentials - physiology; Gaussian noise; Higher order statistics; Humans; least mean square algorithm; Least squares approximation; Least-Squares Analysis; Nervous system; Nervous System Physiological Phenomena; Noise reduction; Normal Distribution; radial basis function network; Radial basis function networks
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2008.2002124

In this study, higher order statistics-based radial basis function network (RBF) was proposed for evoked potentials (EPs). EPs provide useful information on diagnosis of the nervous system. They are time-varying signals typically buried in ongoing EEG, and have to be extracted by special methods. RBF with least mean square (LMS) algorithm is an effective method to extract EPs. However, using LMS algorithm usually encounters gradient noise amplification problem, i.e., its performance is sensitive to the selection of step sizes and additional noise. Higher order statistics technique, which can effectively suppress Gaussian and symmetrically distributed non-Gaussian noises, was used to reduce gradient noise amplification problem on adaptation in this study. Simulations and human experiments were also carried out in this study.