How many channels are needed for MEG?

Channel count in modern MEG systems has been steadily increasing, but are more channels necessary? Assuming that the spatial sampling considerations are satisfied, this question can be answered by examining the MEG system's ability to localize and resolve brain sources. For the simple situation where only uncorrelated sensor noise is present, dipole localization accuracy monotonically increases with increasing number of channels, while for spatially correlated brain noise the accuracy increases only until the number of channels reaches 100 to 200. Beyond this limit the inter-channel separation is comparable to the brain noise correlation distance and increasing the channel count does not help. Contrary to the above dipole result, we show by simulations with up to several thousand channels, that if the data is analyzed by beamformers even in the presence of correlated brain noise, the two-source resolvability and single-source localization accuracy monotonically improve with increasing number of channels. We demonstrate such behavior for a 275 channels system, where we have inserted an artificial dipole into real measured brain noise and resampled the number of channels to 138. Beamformer analysis of the data shows markedly improved localization accuracy when the number of channels is increased from 138 to 275. This finding also signifies that the beamformer performance is not limited by system imperfections when the number of channels is as large as 275. To clarify these results, we illustrate analytically the mechanism of beamformer resolution dependence on the number of channels, using an example of a simple system containing two dipole sources, and uncorrelated sensor noise.