The source estimates are defined on what is called a source space, which specifies the locations of the candidate dipole sources, typically regularly sampled over the cortical mantle or on a volumetric grid. The source space routinely used by MNE is based on the surface defined by the boundary between the gray and the white matter, which consists of a high-resolution mesh with over 100,000 vertices per hemisphere. To reduce the number of dipoles in the source space defined on this surface, it is necessary to decimate the mesh. However, preserving surface topology, spacing, and neighborhood information between neighboring vertices is difficult. Therefore, MNE uses a subsampling strategy that consists of polygon subdivisions using the spherical coordinate system provided by FreeSurfer. For example, an icosahedron subdivided 5 times, abbreviated ico-5, consists of 10242 locations per hemisphere, which leads to an average spacing of 3.1 mm between dipoles (assuming a reasonable surface area of 1000 cm2 per hemisphere), see illustration in Figure 3. The source estimate defined on this low-resolution surface can then be up-sampled and represented on the original high-resolution cortical surface as presented in Figure 4.
