In this talk I will present an approach for the joint estimation of neural source locations, amplitudes and their interactions (functional connectivity) from magnetoencephalographic (MEG) signals. By formulating a state-space for the source locations and their moments, estimation of functional connectivity is reduced to system identification problem in a non-linear state space with a tractable linear sub-structure, whose solution is derived using a Rao-Blackwellized particle smoother (RBPS) combined with an expectation-maximization (EM) algorithm.

This presentation is part of Minisymposium “MS74 - Sequential Monte Carlo methods for inverse estimation in imaging science”
organized by: Narayan Puthanmadam Subramaniyam (Aalto University) , Sara Sommariva (Aalto University) .

Authors:

Narayan Puthanmadam Subramaniyam (Aalto University)

Filip Tronarp (Aalto University, Department of Electrical Engineering and Automation)

Xi Chen (Cavendish Laboratory, University of Cambridge)

Sara Sommariva (Aalto University)

Simo Särkkä (Aalto University, Department of Electrical Engineering and Automation)

Lauri Parkkonen (Aalto University)

Keywords:

bayesian methods, connectivity, expectation-maximization, functional connectivity, image reconstruction, inverse problems, kalman filter, meg, particle filter, sequential monte carlo, statistical inverse estimation methods