| Medical imaging technologies allow the collection of remarkable, three-dimensional pictures of the inside of the body, and have led to noninvasive means of disease diagnosis and treatment planning. However, the proliferation of this technology has resulted in the production of a huge number of medical images, increasing the demand on the radiology work force to a critical level. There is therefore an important need to provide a means of transforming medical image data into information that increases the accuracy and efficiency of radiologists' work. This dissertation focuses on the problem of transforming medical image data in order to provide high-level information about the shapes of anatomical structures. Such information is useful to medical researchers addressing hypotheses relating anatomical shape and pathology, and is also useful to the development of computer-aided diagnosis systems based on shape. This dissertation describes two studies relating shape to pathology of musculoskeletal structures in the shoulder, and uses these studies to motivate research into further interesting questions in shape analysis. Techniques from computational intelligence, such as machine learning, graph matching, feature selection, manifold learning, optimization, and pattern recognition, are used in novel approaches to the steps of a shape analysis pipeline supporting medical research. We describe a machine learning-based approach to eliciting expert knowledge about feature saliency, for use in establishing shape correspondence. We propose a novel approach to medial shape description that localizes shape deformations, and demonstrate a manifold learning-based approach to computing the basic building blocks of it and other medial shape descriptions. Finally, we propose a groupwise paradigm for the computation of a pruning order for the components of medial shape representations, in order to remove unwanted components arising from noise. These contributions to the shape analysis pipeline enable more accurate and intuitively-understood results, enabling medical researchers to gain further understanding into pathological processes in the body. |
