Imaging neural activity using MEG and EEG

The authors have developed a Bayesian framework for image estimation from combined MEG/EEG data. Their results indicate that performance of their imaging approach is superior to that of weighted minimum norm when the image is sparse and focal. Note however that if the image is not sparse, the authors' method would perform poorly since their prior is specifically designed to give sparse focal sources. This observation serves to emphasize the fact that the use of prior information is crucial in extracting useful spatial information from the data. The authors have also found that combining MEG and EEG gives superior results when compared to using the modalities individually. This improvement is due not only to increasing the number of measurements, but also because of the complimentary nature of MEG/EEG. Even when working with the two modalities in combination, significant limitations to electromagnetic imaging exist. Regardless of the number and placement of sensors, reconstructions are generally only reliable if relatively few source clusters exist. If a large number of distributed sources exist, no imaging technique can hope to reconstruct them accurately strictly from the MEG/EEG data given. Such complex distributions will generally be matched as well or better by simpler solutions. Thus, if used on their own, the authors expect MEG/EEG data to be most useful when the number of activated sites is small. Alternatively, when used in combination with fMRI or PET, it may be possible to produce dynamic images of more complex processes.