Particle Filters for Magnetoencephalography

Magnetoencephalography (MEG) is a powerful technique for brain functional studies, which allows investigation of the neural dynamics on a millisecond time-scale. The localization of the neural sources from the measured magnetic fields, based on the solution of an inverse problem, is complicated by several issues. First, the problem is ill-posed: there are infinitely many current distributions explaining a given measurement equally well. Second, the amount of noise on the data is very high, and the main source of noise is the brain itself. Third, the problem is dynamical because the temporal resolution of the data is of the same order of the temporal scale of the neural dynamics. In the last two decades, many different methods have been proposed and applied for solving the MEG inverse problem; however, the search for a reliable yet general and automatic approach to MEG source modeling is still open. Recently we have worked at applying a new class of algorithms, known as particle filters, to the MEG problem. Here we attempt a review of these methods and show encouraging results obtained on both synthetic and experimental MEG data.