Processing of MRI data for simulation and monitoring of drug delivery

The work in this thesis centers around monitoring and simulation of a novel drug delivery system. The major features are the development of a pipeline for creating realistic tetrahedral geometries from MRI data suitable for finite elements simulation, research into methods for quantifying the impact of the shape of an injected drug depot on drug diffusion, and research into automated methods for segmenting hyperplasia growth from MRI images for monitoring the efficacy of the drugs and delivery methods being tested. The error quantification portion of this work exploits assumptions of smoothness in order to employ the stochastic collocation method to study the effect of drug depot shape variability on the outcome of drug diffusion simulations. Simple parameterizations of depot shape are also explored. We demonstrate that once the underlying stochastic process is characterized, quantification of the introduced variability is quite straightforward and provides an important step in the validation and verification process. The hyperplasia segmentation method described here attempts to exploit minimal user interaction to semi-automatically produce an accurate segmentation. The challenge arises from the need to track a combination of strong and weak features near confounding structures. The 3D segmentation process is broken down into a series of constrained 2D segmentation tasks, which are carried out automatically on preprocessed image slices by an active contour model.