Purpose: Minimizing post-operational neurological deficits as a result of brain surgery has been one of the most pertinent endeavours of neurosurgical research. Studies have utilised fMRIs, EEGs and MEGs in order to delineate and establish eloquent areas, however, these methods have not been utilized by the wider neurosurgical community due to a lack of clinical endpoints. We sought to ascertain if there is a correlation between graph theory metrics and the neurosurgical notion of eloquent brain regions. We also wanted to establish which graph theory based nodal centrality measure performs the best in predicting eloquent areas.
Methods: We obtained diffusion neuroimaging data from the Human Connectome Project (HCP) and applied a parcellation scheme to it. This enabled us to construct a weighted adjacency matrix which we then analysed. Our analysis looked at the correlation between PageRank centrality and eloquent areas. We then compared PageRank centrality to eigenvector centrality and degree centrality to see what the best measure of empirical neurosurgical eloquence was.
Results: Areas that are considered neurosurgically eloquent tended to be predicted by high PageRank centrality. By using summary scores for the three nodal centrality measures we found that PageRank centrality best correlated to empirical neurosurgical eloquence.
Conclusion: The notion of eloquent areas is important to neurosurgery and graph theory provides a mathematical framework to predict these areas. PageRank centrality is able to consistently find areas that we consider eloquent. It is able to do so better than eigenvector and degree central measures.
Keywords: Centrality; DSI; DTI; Degree; Diffusion spectrum imaging; Eigenvector; Graph theory; Neuro-oncology; Neurosurgery; Pagerank; Strength.