Automatic Shape Modelling with Applications in Medical Imaging

Abstract: This thesis consists of two parts. The first part is devoted to automatic shape analysis and the second part is devoted to decision support systems in medical imaging. Shape models are widely used in segmentation and shape analysis. The thesis begins with a review of deformable models and the preliminaries of shape modelling. The criteria for Minimum Description Length (MDL) and a new optimization technique for MDL is presented. A new algorithm which uses affine shape to determine parameterisation for curves is also proposed. This part also includes a novel theory for prevention of clustering in correspondence optimization, based on shape variation invariant to curve parameterisation. The first part ends with a chapter on benchmarking algorithms for automatic shape modelling. In the second part of the thesis three decision support systems are described. The first uses a 3D-shape model to segment out the heart in SPECT-images and can be used to diagnose heart infarction. The second system uses a static model for the diagnosis of Parkinson's disease from DatSCAN images. The third system handles the diagnosis of lung-embolie from lung-scint images. This part is concluded with a chapter on segmentation of medical images using 3D active shape models.