Block-Champagne: A novel Bayesian framework for Imaging E/MEG Source

Abstract

Estimating the extents of E/MEG source activities is vital for exploring brain dynamics at high spatiotemporal resolution. However, conventional ESI models exclusively overemphasize on the accurate estimation of the source locations and struggle to reconstruct the extents due to the ill-posed nature of the problem, which requires effective integration of biophysical constraints and proficiency in signal processing methods. In this study, a novel Bayesian framework - Block-Champagne is introduced to provide accurate estimation of extended sources (i.e., both locations and extents). Specifically, a blocksparsity constraint is employed to model the local homogeneity of sources, which can be updated automatically to reconstruct source with arbitrary extents. Furthermore, prior constraints from other modalities, such as fMRI maps, can be incorporated to model interactions between remote sources to further enhance source reconstruction. The performance of Block-Champagne was quantitatively evaluated through simulation experiments, demonstrating its overall superiority under various complex scenarios (i.e., SNR, extent size, & number of extents) compared to benchmark algorithms (including LORETA, EBI-Convex, tsCham, L21-Sissy, & SI-STBF). Moreover, validation results obtained from deep-brain stimulation EEG, epilepsy data, and face-processing multi-modal data further proved the practical feasibility of Block-Champagne. In conclusion, our study reveals the superiority of the proposed BlockChampagne in accurate reconstruction of extended source, positioning Block-Champagne as a highly promising tool for realistic applications where source locations and extents are of equivalent importance.