Robust Bayesian estimation of multiscale brain source activity using MEG data

Robust estimation of multiscale brain source activity has long been a challenge task. The challenge is not only from the reconstruction of thousands of complex brain sources from few hundred sensors, but also the estimation of spatial extent, which is particularly important for the amount of brain tissue that needs to be removed. This paper proposes a scalable Bayesian algorithm that enables better reconstruction of multiscale brain source activity by scaling brain activity during each iteration. Since the proposed algorithm builds upon many of the performance features of the sparse source reconstruction algorithm - Champagne and we refer to this algorithm as MultiScale Champagne (MS_Champ). MS_Champ is able to reconstruct point and cluster sources in multiscale simultaneously. Simulations demonstrate excellent performance of MS_Champ when compared to benchmark algorithms in accurately determining the spatial extent of brain source activity. MS_Champ also accurately reconstructs real MEG data.