Bayesian inference of whole-brain structural connections

Abstract

Structural brain connectivity is used create models of white-matter tract networks within the brain. The location and direction of these structural connections can be estimated by using di usion-weighted imaging data(DWI). Current methods have failed to pro- vide a de nitive model of the structural connectivity of the brain from the data that DWI produces. Previous approaches have often relied on the selection of an arbitrary threshold. In this thesis I provide four di erent generative models based on biophysical characteristics found in the brain. The characteristics examined are e ciency of con- nections, modularity of the brain and similarity across human brains. These models can describe how the structural connectivity and the prior lead to observed and predicted streamline counts from DWI data, which are validated using the predictive power of the generative models. It is shown that using the biophysically motivated characteristics provide a signi cant improvement of predictive power over random models.