In functional magnetic resonance imaging (fMRI) studies, inter-subject anatomical variability of the human brain has been a major challenge in finding reliable functional/anatomical correspondences. Assessment of brain-behavior relations involves a series of geometrical/statistical operations on brain images to minimize such inter-subject variability, so that group maps of brain activity relative to brain anatomy can be developed. Various methods of image registration, segmentation, and analysis have been proposed for mapping functional activity on to anatomical atlases of the brain. The two most common techniques that have been widely accepted and used by neuroimaging scientists are volume-based (VB) analysis using group registration methods and region-of-interest (ROI)-based methods using automated segmentation algorithms or macro/microanatomical probabilistic atlases for labeling. Nevertheless, the analysis results based on these techniques are significantly affected by the accuracy of the selected segmentation and/or registration methods. Furthermore, conventional fMRI data analysis techniques (VB, and ROI-based methods) mainly rely on the assumption that brain processes are common and universal among individual humans; however, besides anatomical differences, there also exist cognitive and behavioral variability among individuals due to differential engagement of brain networks even when performing an identical cognitive task. In this thesis, I have assessed the impact of anatomy-based alignment techniques (VB, and ROI-based methods) on sensitivity of fMRI data group analysis. I evaluated the effect of the type of inter-subject registration used and related factors on sensitivity of group-level fMRI data analysis. Furthermore, I have also assessed the goodness of fit of probabilistic maps by proposing an evidence-based framework for evaluation of probabilistic maps. As a test model, I have selected the human auditory cortex. Auditory cortex is an interesting yet challenging case with substantial inter-individual functional/anatomical variability. For the sake of ROI-based method of analysis, I have proposed a novel approach for automatic segmentation of Heschl's gyrus, which is the landmark for primary auditory cortex. Finally, in order to assess the impact of inter-subject variability in anatomy on functional organization, I analyze data from an fMRI study, which demonstrates that the degree to which anatomical registration compensates for functional variability depends on the brain region activated. / Thesis (Ph.D, Computing) -- Queen's University, 2010-04-29 07:07:55.77

The underlying causes of variability in the electrical activity of hearts from individuals of the same species are not well understood. Understanding this variability is important to enable prediction of the response of individual hearts to diseases and therapies. Current experimental and computational methods for investigating the behaviour of the heart do not incorporate biological variation between individuals. In experimental studies, experimental results are averaged together to control errors and determine the average behaviour of the studied organism. In computational studies, averaged experimental data is usually used to develop models, and these models therefore represent a 'typical' organism, with all information on variability within the species having been lost. In this thesis we develop a methodology for modelling variability between individuals of the same species in cardiac cellular electrophysiology, motivated by the inability of traditional computational modelling approaches to capture experimental variability. A first study is conducted using traditional modelling approaches to investigate potentially pro-arrhythmic abnormalities in rabbit Purkinje fibres. A comparison with experimental recordings highlights their wide variability and the inability of existing computer modelling approaches to capture it. This leads to the development of a novel methodology that integrates the variability observed in experimental data with computational modelling and simulation, by building experimentally-calibrated populations of computational models, that collectively span the variability seen in experimental data. We apply this methodology to construct a population of rabbit Purkinje cell models. We show that our population of models can quantitatively predict the range of responses, not just the average response, to application of the potassium channel blocking drug dofetilide. This demonstrates an important potential application of our methodology, for predicting pro-arrhythmic drug effects in safety pharmacology. We then analyse a data set of experimental recordings from human ventricular tissue preparations, and use this data to develop a population of human ventricular cell models. We apply this population to study how variability between individuals alters the susceptibility of cardiac cells to developing drug-induced repolarisation abnormalities. These abnormalities can increase the chance of fatal arrhythmias, but the mechanisms that determine individual susceptibility are not well-understood.