| The main research backgrounds of this thesis are the project: " cardiac MRI image analysis and visualization", supported by the HongKong (HK) special government grants council and cooperated with the computer department of the university of HK Chinese, and the project: " microscopic image segmentation and recognition on the cancer calls fallen into peritoneal effusion", supported by the computer vision lab in department of computer of Nanjing university of science & technology.At present, Magnetic Resonance Imaging (MRI) technology has become an important tool in clinic diagnosis to the human Cardiovascular Diseases (CVD). As a powerful biologic pump to circulate the human blood, the Left Ventricle (LV)'s dynamic behavior is very important to CVD examination. So, in the analysis to the cardiac MRI images, a set of spatial sequential images sampled during a heartbeat, the research keys are concentrated on the movement and the deformation of the LV, including: (1) LV segmentation;(2) LV movement reconstruction;(3) LV stress analysis;(4) data visualization. The key therein mentioned above, that is LV segmentation, is the first step and strongly affects the later analysis. However, due to the anatomy construction complexity, the LV boundary in cardiac MRI image is badly blurry, and the low gradient weak boundary and the artifact often can be found in LV region. Because of the reasons described above, it is difficult to have good segmentation if using the traditional segmentation method. The active contour model, that is Snake model presented by the Kass et al., is a new mutual image segmentation tool. But because only the local information of image is used, the classical Snake is very sensitive to the original contour and can not to treat the topology change during the contour deformation. These shortages lead to the Snake not to meet the requirement from cardiac MRI image segmentation.The cancer is one of the most fatal diseases around the world at present. The correct diagnosis of the initial stages is a key to heal this disease. It is one of the most important and difficult tasks of the pathologists to detect and diagnose cancer cells fallen into peritoneal effusion. Some of these cancer cells may not be detected due to the fact that they may be similar to non-cancer cells in shape or size, such as the similar shape of mesothelial cells when heteromorphism occurs in their nuclei, or the low quality of the images, or the pathologists' different experiences and variable decision criteria. Due to a variety of reasons described above, the false-positive or false-negative ratio with which pathologists fail to diagnose cancer cells fallen into peritoneal effusion is not low at present. So, it isvery significant to correctly recognize the cancer cells by combined the moderncomputer image processing technology and the artificial intelligence technology with the pathologists' experiences.Based on the backgrounds mentioned above and the works of the former researches, this thesis presented some innovation researches in four topics as follows: 1.In order to overcome the shortages of the classical Snake model, such as final contour being sensitive to original contour and the computing time being rather long, et al., an improved Snake model is presented. Besides area energy function, a new elasticity energy function and a new rigidity energy function are introduced in this model. And a two-phases Greedy optimization algorithm are proposed as well as other several auxiliary algorithms, including the algorithm for to adaptively increase or to adaptively decrease the Snake's vertexes and the algorithm for to automatically choice positive or negative sign of the coefficient of the area energy, et al. The segmentation tests on the vivo cardiac MRI images show that the improved model can efficiently get over the classical Snake's lacks mentioned above.2. In allusion to the classical Snake model being unable to treat contour topology change and numerical method often being unsteady, a cardiac MRI image segmentat...
|
PDF Full Text Request