SPACE-TIME SPARSITY REGULARIZATION FOR THE MAGNETOENCEPHALOGRAPHY INVERSE PROBLEM

The concept of "space-time sparsity" (STS) penalization is introduced for solving the magnetoencephalography (MEG) inverse problem. The STS approach assumes that events of interest occur on localized areas of the cortex over a limited time duration, and that only a few events of interest occur during a measurement period (or epoch). Cortical activity is reconstructed by minimizing a cost function which fits the data with a sparse set of space-time events using a novel expectation-maximization (EM) algorithm. We employ spatial and temporal basis functions to reduce the dimension of the data fitting problem and combat noise. Simulations suggest that our approach could be useful for identifying sequential relationships in the brain