Coupled non-parametric shape and moment-based inter-shape pose priors for multiple basal ganglia structure segmentation

Brain tissue and structure segmentation in magnetic resonance (MR) images is a fundamental problem in clinical studies of brain structure and function. Due to limitations such as low contrast, partial volume effects, and field inhomogeneities, the delineation of subcortical (basal ganglia) structures such as caudate nucleus, putamen, and thalamus from white matter, gray matter and cerebrospinal fluid (CSF) is a very challenging problem. This thesis presents a new method for simultaneous segmentation of multiple brain structures. We formulate the segmentation problem as a maximum a posteriori estimation problem, in which we incorporate statistical prior models on the shapes and relative poses of the structures of interest. Our method is motivated by the observation that neighboring or coupling structures in medical images generate configurations and co-dependencies which could potentially aid in segmentation if properly exploited. Our coupled shape priors are learned through nonparametric multivariate kernel density estimation based on training data. Relative pose priors are modeled via standard moments. Given this framework, the segmentation problems turns into an optimization problem, which we solve using active contours. We present experimental results on synthetic data as well as on a rich set of real MR images demonstrating the effectiveness of the proposed method in segmenting basal ganglia structures as well as improvements it provides over existing approaches