Modeling Connectivity and Spatiotemporal Hemodynamics to Improve fMRI Analysis

The brain and its activity are difficult to observe directly in living subjects, and one of the most versatile tools developed to overcome this is functional magnetic resonance imaging (fMRI). This allows for the structure and functioning of the brain to be imaged in real time, which has made it integral to many studies of brain function. However, fMRI does not measure the neuronal activity directly, but the BOLD (blood oxygen level dependent) signal induced by changes in blood flow and oxygenation when that activity occurs. Thus, it is important to understand how the BOLD signal is generated from neural inputs. This has been explored extensively temporally, but only recently have the spatial effects related to the BOLD signal been studied, including how it varies through the depth of the cerebral cortex. In addition, once the activity has been determined, many studies have attempted to use it to infer connections within the brain by examining correlations in the data. It is generally assumed that areas of the brain strongly connected to each other will exhibit strong correlations in their BOLD signals. From this, a map of the brain is developed by subdividing the brain into regions and determining the connections between them. This is then thresholded into a graph and analyzed, and conclusions are drawn about the brain from its properties. However, this is done without determining whether the networks used for comparison to draw these conclusions are similar enough in structure to the brain to provide a valid comparison. This thesis addresses both of the above issues; first, it further develops a previous spatiotemporal model for the BOLD response, extending it to incorporate a previously neglected nonlinearity, and for it to fully model how the BOLD response varies with cortical depth. Secondly, it examines key assumptions used in many studies of brain connectivity, and whether or not they are correct