| Traditional practice in neuroscience has been to examine the brain in terms of isolated components extracted from images. However, more recent trends have moved towards the examination of the entire brain in order to observe the complete topology and to capture emergent behavior not present at the component level. Network based models enable us to study how low level interactions in the brain can produce emergent behaviors, and identify regions that are most central to those behaviors. This project takes two approaches to the understanding of the functional brain as a network.;First, there is evidence of the existence of critical nodes in self-organized networks, such as the brain, that are essential to information flow. Centrality is a class of metrics that attempts to identify such nodes. In this work, we show that a new centrality measure, leverage centrality, is more effective at identifying these critical nodes in the brain than other centrality measures. The role of high centrality nodes was further investigated through network attack studies. In these experiments, we studied the effect of random failure of nodes or targeted attack of highly central hubs on both network structure and dynamics. The findings of this work demonstrated that the human functional brain network is in fact highly resilient to both types of attack. In terms of structural impact, the functional brain networks maintained significantly more efficient connectivity than equivalent random networks. Furthermore, dynamical simulations demonstrated that the ability to transfer information throughout the network remained intact.;The second approach to studying the brain as a network employs agent based modeling techniques. Agent based models have been shown to be extremely effective at capturing emergent behavior arising from complex networks. A computerized agent based model was developed to represent the functional brain network, and we demonstrate that this model can produce highly variable and complex behaviors and is capable of supporting computation. Since this model is capable of producing such a wide array of behaviors, we employ genetic algorithms to tune the model parameters in order to produce desired behaviors. We discuss several genetic algorithm designs and their utility given particular problem constraints. |
