Criteria for fitting MEG dipoles with fMRI position constraints

Since MEG and EEG inverse problems are "illposed" (i.e., solutions cannot be uniquely determined), additional information such as fMRI and PET activation data can be used to impose additional constraints [1]-[6]. For example, fMRI data were used to fix the locations of equivalent current dipoles at activation areas [1][2][5] or to give higher weights for finding dipoles at activation areas [3][4][6]. The latter method is capable of obtaining dipoles not only at fMRI activation areas but also at areas where fMRI fails to show activation. It means, however, that the nonuniqueness nature of the inverse solutions remains in this method, and that flexibility in finding dipoles and uniqueness of solutions are a tradeoff to be optimized. In comparison, the former method of fixing dipole locations shows active dipoles only at fMRI activation areas, but has the advantage that the solutions are uniquely soluble by a least-squares method with no additional assumptions. This method has been applied to the analysis of actual data [5]. However, there are two issues that need to be clarified further: (1) how to place discrete dipoles for approximating spatially extended neural sources, and (2) how to account for influences on neighboring dipoles, or "crosstalk" when plural dipoles are fixed in a small neighborhood. Although average behaviors were previously analyzed using Monte Carlo simulations [3], their spatially anisotropic characteristics need to be investigated for effective application of the method. In the present work, we carried out simulations of solving for dipoles with fixed positions (constrained dipoles) to clarify the above two issues, and we applied the obtained criteria to actual data.