The analysis of functional networks is a multi-step procedure, in which methodological choices and assumptions must be made. The choice of the post-processing techniques such as artifact reduction, filtering, normalization, and nuisance variable regression can influence the results. Both ICA and seed-based analysis of fMRI data have technical and practical limitations that remain to be addressed and have been outlined in a recent review.135 Similarly, graph theoretical network investigation requires methodological decisions that can bias outcomes and conclusions. For example, appropriate statistical thresholding for network definition and extraction remains a critical issue for this approach.14 In addition, it is important to recognize that the spatial resolution of present EEG/MEG recording techniques poses limitations on the measurement of deep brain neuronal activity and therefore on the interpretation of the results.33 Finally, data about the sensitivity, specificity and reliability of task-free fMRI and EEG/MEG data are still limited.136 However, despite these important limitations, recent brain connectivity studies using different recording techniques and analytical approaches show converging results,137 suggesting that a more cohesive view of brain (dys)function in dementia may arise from the study of networks.
