Evaluation of Brain Source Separation for MEG Data Applying JADE, Fast-ICA and Natural Gradient-Based Algorithm with Robust Pre-Whitening Technique

Autor:	Mamoru Tanaka, Takayuki Arai, Jianting Cao, Tsunehiro Takeda, Yoshio Konno, Hiroshi Endo
Rok vydání:	2004
Předmět:	business.industry Computer science JADE (programming language) Pattern recognition Covariance Independent component analysis Power (physics) Pre whitening Source separation Artificial intelligence Natural gradient business computer Algorithm computer.programming_language Fast ica
Zdroj:	Journal of Signal Processing. 8:461-472
ISSN:	1880-1013 1342-6230
DOI:	10.2299/jsp.8.461
Popis:	Independent component analysis (ICA) has been applied to magnetoencephalographic (MEG) data to determine the behavior and localization of brain sources. In this work, both nonaveraged single-trial data and averaged multiple-trial data are analyzed, in order to study the relationship between the performance of decomposition and the number of averages across data trials. To evaluate the performance of source decomposition, 1) the ratio of source decomposition, which indicates whether source components will be decomposed or not, 2) the power of decomposed components, which is calculated as the covariance of decomposed components, and 3) the accuracy of source estimation, are demonstrated. In addition, a number of existing ICA algorithms such as JADE, Fast-ICA, and the natural gradient-based algorithm with a robust pre-whitening technique are used to decompose MEG data. Our results show the relationship between the accuracy of source decomposition and the number of averages and, by applying the ICA approach, the number of averages can be reduced.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::5371c70fd84b5f99168c05941b789b30 https://doi.org/10.2299/jsp.8.461 Zobrazit plný text záznamu