Applying ICA in EEG: Choice of the Window Length and of the Decorrelation Method - Biomedical Engineering Systems and Technologies - page 261

be reliably applied. In particular, for dipolar mixtures and using sphering as initializa-
tion, AMICA showed impressive performances with very low data length even for high
number of channels: 3 s of data length (512 Hz sampling rate) are sufficient to get an ex-
cellent PI below 0 . 03 for the separation of 48 sources. The main drawback of AMICA
is its time consumption: it requests more than 60 seconds for this particular example,
while FastICA converges in less than 1 s to get similar PI , although needing 7 s of data.
An immediate perspective to this work would be to use more realistic time-structured
data, obtained using only modelled neural sources and realistic mixtures (head models).
Besides confirming our conclusions for the studied algorithms, this type of simulation
setup would allow the evaluation of second order statistics BSS algorithms (SOBI and
similar, also widely used for EEG analysis). It might be also useful if algorithms could
be tested on more data channels, in order to asses their performances in the context of
high-resolution EEG (more than 64 channels).
An interesting perspective, for the specific case of EEG source separation, is to
consider new contrasts for BSS algorithms, balancing between dipolarity and indepen-
dence. Indeed, source independence is known to be often unrealistic for EEGs, as strong
synchrony is very likely to appear between distant areas in the brain. A relaxation of the
independence constraint might then enhance the EEG source separation performance.
References
1. Schomer, D., Lopes da Silva, F. (eds.): Niedermeyers's Electroenephalography: Basic Prin-
ciples, Clinical Applications and Related Fields. Wolters Kluwer, Lippincott Willimas &
Wilkins (2011)
2. Sanei, S., Chambers, J.: EEG Signal Processing. John Wiley & Sons (2007)
3. Scherg, M., Berg, P.: Use of prior knowledge in brain electromagnetic source analysis. Brain
Topography 4, 143-150 (1991)
4. Cichocki, A., Amari, S.: Adaptive Blind Signal and Image Processing Learning Algorithms
and Applications. John Wiley & Sons, New York (2002)
5. Delorme, A., Palmer, J., Onton, J., Oostenveld, R., Makeig, S.: Independent eeg sources are
dipolar. PLoS ONE 7(2), e30135 (2012)
6. Hyvärinen, A.: Fast and robust fixed-point algorithms for independent component analysis.
IEEE Transactions on Neural Networks 10(3), 626-634 (1999)
7. Bell, A.J., Sejnowski, T.J.: An information-maximization approach to blind separation and
blind deconvolution. Neural Computation 7, 1129-1159 (1995)
8. Palmer, J., Makeig, S., Delgado, K., Rao, B.: Newton method for the ICA mixture model. In:
IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2008,
March 31-April 4, pp. 1805-1808 (2008)
9. Cardoso, J.: High-order contrasts for independent component analysis. Neural Computa-
tion 11(1), 157-192 (1999)
10. Särelä, J., Vigário, R.: Overlearning in marginal distribution-based ICA: analysis and solu-
tions. J. Mach. Learn. Res. 4, 1447-1469 (2003)
11. Onton, J., Makeig, S.: Information-based modeling of event-related brain dynamics. Progress
in Brain Research 159, 99-120 (2006)
12. Delorme, A., Makeig, S.: EEGLab: an open source toolbox for analysis of single-trial
EEG dynamics including independent component analysis. Journal of Neuroscience Meth-
ods 134(1), 9-21 (2004)