Spatio-temporal EEG brain imaging based on reduced Kalman filtering

The non-invasive EEG neuronal activity estimation is still an open research area, because the limited number of sensors and the thousands of possible sources make it an ill-posed inverse problem. In recent years several authors have included temporal information on the EEG inverse problem in order to reduce its dimensionality, followed by Kalman filters in order to fix the lack of information. But these approaches are impractical with large scale matrices of highly dense grids of dipoles filling the brain. On this paper a new reduction methodology for the Kalman filter estimation is proposed, it allows efficient computations without losing robustness and stability. The validation of the proposed methodology was performed in a realistic BEM head model with dipoles restricted to the gray matter and oriented perpendicular to the cortical surface. A well known database provided by the SPM8 software was implemented for comparison purposes with a LORETA-like smoother. These results show that the proposed methodology effectively improves the neuronal source reconstruction and localization, without increasing the computational cost.