eduzhai > Applied Sciences > Engineering >

Spectral independent component analysis with noise modeling for M/EEG source separation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 26 pages

Abstract: Background: Independent Component Analysis (ICA) is a widespread tool forexploration and denoising of electroencephalography (EEG) ormagnetoencephalography (MEG) signals. In its most common formulation, ICAassumes that the signal matrix is a noiseless linear mixture of independentsources that are assumed non-Gaussian. A limitation is that it enforces toestimate as many sources as sensors or to rely on a detrimental PCA step.Methods: We present the Spectral Matching ICA (SMICA) model. Signals aremodelled as a linear mixing of independent sources corrupted by additive noise,where sources and the noise are stationary Gaussian time series. Thanks to theGaussian assumption, the negative log-likelihood has a simple expression as asum of divergences between the empirical spectral covariance matrices of thesignals and those predicted by the model. The model parameters can then beestimated by the expectation-maximization (EM) algorithm.Results: Experiments on phantom MEG datasets show that SMICA can recoverdipole locations more precisely than usual ICA algorithms or Maxwell filteringwhen the dipole amplitude is low. Experiments on EEG datasets show that SMICAidentifies a source subspace which contains sources that have less pairwisemutual information, and are better explained by the projection of a singledipole on the scalp.Comparison with existing methods: Noiseless ICA models lead to degeneratelikelihood when there are fewer sources than sensors, while SMICA succeedswithout resorting to prior dimension reduction.Conclusions: SMICA is a promising alternative to other noiseless ICA modelsbased on non-Gaussian assumptions.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×