Comparison and Application of Selected Statistical Shape Models in Medical Imaging

This paper reviews several kinds of 2D shape representations by a set of parameters based on labeled points, Fourier descriptors and wavelet descriptors, resp.. Shape models are derived by statistical analysis of parameters corresponding to a set of example shapes. Each model consists of a parameter vector describing mean shape and a set of modes of variation for parameters characterizing shape variability. Seven shape models, some of them differing in parameter normalization, for axial slices of spinal vertebra are compared with respect both to their compactness in parameter space and to their scope in corresponding space of shapes. A model based method for segmenting 2D gray level images is developed by formulating boundary finding as an optimization problem with respect to parameters varying according to the modes of variation. Our method includes an easy and fast interactive improvement of segmentation outcome.