Dynamic Bayesian imaging using the magnetoencephalogram

We describe a new approach to imaging neuronal current sources from magnetoencephalogram (MEG) measurements associated with sensory, motor or cognitive brain activation. Previous approaches use weighted minimum norm inverse methods which produce spatially smooth solutions. These results are inconsistent with functional activation studies using fMRI or PET, which reveal a sparse localized nature of activation in the cerebral cortex. We use a Bayesian technique with a Gibbs prior reflecting this expectation. The prior, combined with a Gaussian likelihood function, forms the posterior density, which we can maximize to produce a non-linear estimate of the primary neural current field. We also investigate marginalizing out the amplitude time-series, and compare the joint and marginal MAP estimates. We apply the methods to phantom data and show favorable performance in comparison to minimum norm approaches