An fMRI-MEG Integrative Neuroimaging Method: Improvements of its Accuracy and Robustness by Suppression of fMRI-invisible Coherent Activities

Abstract

One of possible problems in fMRI-MEG integrative analysis is mismatches between activated regions detected by fMRI and MEG. These mismatches cause serious degradation of estimation accuracy, especially in the case that fMRI-invisible activities have high temporal correlations to activities detected by fMRI. We proposed a spatial filter which can achieve the accurate reconstruction of MEG source activities even in the case that a priori information of fMRI is insufficient. The proposed filter is based on generalized least square (GLS) estimation method. The GLS method requires to determine the noise covariance matrices, and the proposed filter utilizes measured MEGs for the determination. In the present study, principal component analysis is applied to the measured MEGs to determine the noise covariance matrices. Simulation results with conditions that fMRI-invisible MEG sources are present demonstrated that the proposed filter could reconstruct MEG source activities more accurately than the methods based on both ordinary least square method and minimum variance beamformer. The validities of the proposed method were also discussed, with measured data from the experiment using an apparent motion visual stimulus. The results demonstrated that the proposed method could reconstruct reasonable time courses of activations.