Bayesian EEG Dipole Source Localization using SA-RJMCMC on Realistic Head Model

Abstract

In this study, electroencephalography (EEG) inverse problem is formulated using Bayesian inference. The posterior probability distribution of current sources is sampled by Markov Chain Monte Carlo (MCMC) methods. Sampling algorithm is designed by combining Reversible Jump (RJ) which permits trans-dimensional iterations and Simulated Annealing (SA), a heuristic to escape from local optima. Two different approaches to EEG inverse problem, Equivalent Current Dipole (ECD) and Distributed Linear Imaging (DLI) are combined in terms of probability. EEG inverse problem is solved with this probabilistic approach using simulated data on a realistic head model. Localization errors are computed. Comparing to Multiple Signal Classification algorithm (MUSIC) and Low-Resolution Electromagnetic Tomography (LORETA), using MCMC methods with a Bayesian approach is useful for solving the EEG inverse problem.

Human scalp EEG processing: Various soft computing approaches

“…Source reconstruction has been performed under suitable spatial and temporal constraints estimated by Bayesian method. EEG source reconstruction was done in [258] according to both ECD and DS models by formulating the inverse problem as Bayesian inference, like in (1). The forward model was constructed by Markov chain Monte Carlo (MCMC) method.…”

“…Bayesian linear discriminant analysis has been applied for EEG classification in BCI in [269], with a superior performance than SVM and linear discriminant. Adopted from [258].…”