[Electroencephalography inverse problem by subspace decomposition of the fourth-order cumulant matrix].

It is an important topic in electroencephalography (EEG) research to localize the EEG activity sources from the scalp recordings. In this paper, based on the fourth-order cumulant matrix, a new sub-space decomposition algorithm is proposed for the EEG inverse problem. As the second-order moments (cumulants) has the drawback of being sensitive to the noise covariance. Using the fourth-order cumulants we need not know the noise covariances, as long as the noise is Gaussian. Computer simulation study on a three-layer concentric sphere head model shows its better performance than the two-order cumulate method in depressing the spatial coherent Gaussian noise.