A dynamic solution to the ill-conditioned magnetoencephalography (MEG) source localization problem

Dynamic inverse solutions are described that localize dynamic brain activity measured from extraneous magnetic fields at an array of sensor positions on the scalp. The measured spatiotemporal magnetic fields are produced by small but coherent neuronal currents and are recorded using arrays of multichannel SQUID (superconducting quantum interference devices) gradiometers. We outline inverse mapping procedures that combine models of current dipoles with dynamic state-space estimation algorithms. These algorithms are based on a series of hierarchical mathematical relationships that employ principles from electrical and electromagnetic field theory to relate the observed MEG measurements to the current source dipoles. As we show, one solution that is naturally suited to this type of problem is the well-known Kalman filter and smoother