Reconstruction of current dipoles based on tensor decomposition of multipole coefficients

Abstract

This paper presents a novel method for solving magnetoencephalography (MEG) inverse problems using the CANDECOMP/PARAFAC decomposition of a tensor in which Hankel matrices composed of the multipole moments of the observed data are aligned. The proposed method can reconstruct the positions and moments of the current dipoles in the human brain projected on the $xy$-plane more stably than the conventional direct method using Prony's algorithm. A novel spherical-spline-based method for the computation of the multipole moments is also proposed.