Deformable Model Analysis And Research On Its Application In Cardiac MR Images

Currently, the deformable models have become an important tool of the medical image analysis. It incorporates the data constraints derived from images and the prior knowledge about the Region Of Interest (ROI) into the variational framework and its application has covered image denoising, image segmentation, image registration, image inpainting, surface reconstruction and motion tracking, etc. In this paper, the parameter active contour model, geometric active contour model and deformable surface model are analyzed and some research work on its application in the segmentation of the cardiac MR images and the surface construction of the left ventricle are presented.When the classical parametric active contour model(Snake) is applied to segment the image, an initial contour must be set near the boundary of ROI and the model cannot segment deeply concave regions accurately. On the basis of analyzing the Fast Snake model, an improved fast parametric active contour model is proposed. With the result of analyzing the Greedy Algorithm, an local area energy term is defined and added to the target function for capturing the image feature in a wider region and guiding the deforming curve into the concave regions. The proposed model has no special requirement to the location and shape of the initial contour. The encouraging effectiveness of the model is demonstrated with the experiments on the left ventricle MR images.Due to weak edges and the low contrast with the surrounding tissues, the deforming curve will leak from the boundary of the left ventricle when the snake model is applied. An Snake model with shape constraints is proposed to address the difficulty. After the train image samples have been aligned and the variation modes have been analyzed, the shape allowable space of the left ventricle is constructed. According to the properties of the cardiac MRI images, the shape constraint energy field around the average shape is created by Fast Marching Method(FMM).After the shape constraint energy term is added, the Snake model can prevent the deforming curve from leaking out of the low contrast regions effectively. The deformed curve is subject to the shape constraints by mapping it to the shape allowable space. The segmentation experiments demonstrate the effectiveness of the proposed model.Due to cardiac deformation and blood flowing, weak edges, local gradient maximum regions and artifacts are found in the MR images. The difficulties are analyzed in this paper when the level set method is applied to segment those MR images . A two-stage algorithm is thus proposed: first, incorporating prior knowledge about cardiac MR images and image histogram, the fuzzy C-means clustering is applied and with the clustering results, the velocity function of the pixels in the narrow band is constructed to search for the rough boundary of left ventricle. Secondly, another velocity function based on Gradient Vector Flows(GVF) is designed to locate the edges accurately. The experimental results of the cardiac MR images demonstrate the practical value of the proposed model. Others, a distant function template method(FTM) is proposed to reinitialize the narrow band. By comparing the FTM with FFM, it shows that the FTM can improve the speed of segmenting the images on the condition of the same evolving velocity.According to the property of the cardiac MR images, an image segmentation algorithm is proposed by unifying region statistic model and image gradient information. By incorporating prior knowledge about cardiac MR images and the image histogram, the numbers of the Gauss components are determined and the parameters of the components are estimated by Maximum-likelihood principle and Expectation Maximum algorithm. With the pixel posteriori probability belonging to ROI, the velocity function of Level Set is constructed to search for the rough boundary of ROI. Then ,another velocity function based on the gradient intensity is designed to locate the true boundary. The results of the experiments demonstrate the proposed algorithm can...