Sparse reconstruction by separable approximation
IEEE Transactions on Signal Processing
Hi-index | 0.00 |
We examine the performance of "Space-Time Sparsity" (STS) penalized reconstruction of brain activity from magneto-/electroencephalographic (MEG/EEG) recordings. We propose two STS priors, both of which favor activation of a few localized areas of cortex over a limited time duration and bandwidth via appropriate basis functions. This provides a reasonable model of true brain activity which minimizes the impact of the inherently ill-conditioned, low SNR, spatial inverse problem. We use an expectation-maximization (EM) algorithm to solve the STS penalized least-squares cost function. The solution localizes brain activity in space and time, providing support for a refined signal estimate (e.g. minimum norm least-squares). We illustrate the approach on both simulated and real data and provide preliminary theoretical analysis of performance.