Spatial-Temporal Source Reconstruction of MEG via Variational EM Algorithm

Magnetoencephalography(MEG) is a new noninvasive brain imaging technique reconstructed electronic activities of brain by measured the magnetic field surrounding scalp. The aim of this paper is to explore a new method of MEG source spatio-temporal reconstruction based on optimizing the reconstructed MEG source model. We make the assumption that the stimulated electronic activities of the brain are located on a particular part of cortex where we partition it with multiple even voxels. In terms of Biot-Savart Law of electromagnetism, the spatial source model is built with multiple unknown parameters which reflect the information of the source location. Then, we try to solve this parameters optimization as a Maximum-likelihood estimation (MLE) using variational EM algorithm. According to the application of this approach, this paper also addresses that the solution of MEG signal reconstruction should be considered to avoid overlapping the calculation complexity, which may result in too expensive calculation to practice. Whereas, this approach also provides a new possibility and the new angle to solve MEG source reconstruction.