Non-Linear Iterative Regularization Algorithm for Neural Activity Reconstruction from EEG Data

Abstract

In this work a novel non-linear iterative regularization algorithm applied to the reconstruction of neuronal activity is presented. A physiologically-based non-linear spatio-temporal constraint is used for solving the dynamic inverse problem associated to the reconstruction of neural activity of the distributed sources. The proposed method includes the spatio-temporal constraint in a cost function based on a $l$ -2 norm. A simulated EEG data-set is used in order to evaluate the performance of the proposed algorithm for three and five simultaneously active sources under several signal-to-noise ratios, by using relative error measurement. A comparison analysis is performed against the MSP and IRA-L2 source reconstruction methods, where the proposed non-linear method improves the reconstruction of neural activity in terms of the relative error.