A finite element model is more correct. Even a homogenous sphere may be acceptable. The big issue is that many contacts are in VM and “see” a very different EEG than contacts in GM. Somehow you have to take this into account. In GM you may have very focal activities that even may be uncorrelated to other activities in other parts of GM. I can’t see how you can do source reconstructions on these. Using only VM contacts may be a way for source localization. Connectivity measures should work in both VM and GM contacts