EEG Source Estimation using Multiple Unscented Particle Filtering

In this paper, we propose a multiple unscented particle filters approach to reconstruct the dynamic brain source localization. Localization of the brain sources that generate Electroencephalographs (EEG) has been a challenging problem in research, clinical and technological applications associated with the brain. The active areas in the brain are designed as equivalent current dipoles, and the moments and positions of these dipoles are the target to be estimated. We are formulating this EEG dynamic brain source localization problem as a nonlinear state-space model. Nowadays, particle filters (PF) represent the optimality in estimating the nonlinear and non-Gaussian systems. However, PF are inefficient in high-dimensional state-spaces because the number of particles desired increases super-exponentially with the dimension of the state. Unscented particle filter (UPF) has been emerged recently to improve the performance of PF. Therefore, we introduce a newly Multiple Unscented Particle Filter (MUPF) to deal with the problems that have high-dimensional spaces. In addition, we propose the MUPF algorithm to estimate the positions and the moments of the dipoles over time. Simulation results on synthetic and real EEG data using the MUPF provide effective tracking for dipole localization compared to the standard PF, multiple PF and UPF.