Construction Of A Three-dimensional Statistical Shape Model Of Lung CT Images

Medical image segmentation has a very important significance in biomedical imaging, clinical diagnosis and pathology analysis. It is the base of image processing, image analyzing and understanding. Because of the complexity of the human anatomy and the irregularities of organ shapes by individually, just using grey-level depended method is not able to reach a robust segmentation result. Since the introduction of the prior knowledge of target shapes, the statistical shape model (SSM) based methods are widely applied in medical image processing domains. However, building a high-quality shape model is the key to implement SSMs.We summarize three steps to build a SSM. At the first, the surface data set is collected from medical images. And then the landmark points on the surfaces are corresponded across the whole data set. The statistical shape model is generated by statistical analysis of the positions of landmarks. In this thesis, we propose a spherical conformal mapping based parametrization method to complete the automatically landmarks correspondence. First of all, an improved region growing method is proposed to get the pure surfaces from there-dimensional medical images.Then, spherical conformal mapping method is utilized to implement the surface parameterization, by conformal mapping of three constraints and zero centroid constraints affect access model was produced for comparison, we know the model-specific capabilities acquired three constraints significantly better than zero centroid constraints; Finally approximation algorithm to achieve a rigid registration training model sample set iterative point.We use a statistical model by 86 lung samples constructed versatility and specificity assessment, indicating that with the number of samples increases, the model will has stronger versatility; and better specificity.