Functionally-structured Bayesian model for localizing neural activity and information in magnetoencephalography signals

Abstract

Magnetoencephalography (MEG) is a noninvasive method that can measure human brain activity with high temporal resolution. However, the spatial resolution of MEG is not sufficient to reveal neural mechanisms. Although MEG source estimation overcomes this problem to some extent, the combination of MEG source estimation and multivariate analysis results in "information spreading", where significant predictions are observed in brain areas outside the true signal source location. This paper describes a model that achieves high source estimation accuracy and suppresses information spreading simultaneously. The proposed approach is based on a Bayesian estimation model that utilizes functional structure of the human brain. We compare the performance of the proposed model with simulated data generated under various signal-to-noise ratio conditions. The results show that the functionally-structured Bayesian model achieves source estimation accuracy that is better than that of conventional source estimation models. Additionally, the comparison of information spreading among the models reveals that our model outperforms the conventional ones. These results suggest that information spreading in the MEG source estimation can be suppressed while maintaining high source estimation accuracy.