Bayesian tracking and multi-core beamforming for estimation of correlated brain sources

The main contribution of this paper is the general framework, termed multi-core Beamformer Particle Filter (multi-core BPF), for solving the ill-posed EEG inverse problem. The method combines a particle filter (statistical approach) for reconstruction of the brain source spatial locations and a multi-core Beamformer (deterministic approach) for estimation of the corresponding dipole waveforms in a recursive way The intuition behind is to benefit from the advantages of both deterministic and statistical inverse problem solvers in order to improve the estimation accuracy without increasing the complexity and the computational cost. Our simulations show that the proposed algorithm can reconstruct reliably the few most active (the dominant) brain sources that have generated the registered EEG measurements. The main advantage of the method is that in contrast to conventional (single-core) Beamforming spatial filters, the proposed Multi-core Beamformer explicitly takes into consideration the potential temporal correlation between the dipoles.