基于水平集方法的三维超声颈动脉斑块分割-生物医学工程专业论文.docxVIP

华 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract Automatic segmentation of the carotid plaques from 3D ultrasound images has been shown to be an important task for monitoring progression and regression of carotid atherosclerosis. Considering the complex structure and echo morphology of plaques, a fully automatic segmentation method based on media-adventitia (MAB) and lumen-intima (LIB) boundary priors is proposed, so that the segmentation of plaques is converted to a task to differentiate normal vessel wall and plaque complex. This method mainly contains two steps: contour initialization and level-set evolution. The first step is to detect the initial points using the grey level information between MAB and LIB, and then the structure information is incorporated with the image information into the level-set framework. The final segmented contour in the level-set process is the plaque “outer” contour and the region between this “outer” contour and LIB is defined as plaque region. Besides, we obtain the manual segmentations from two sonographers at different time point, and the mean contours are computed, which are taken as the ground truths and used to examine the validity and accuracy of the proposed segmentation algorithm. The optimal values of the segmentation parameters are determined using the Evolutionary Strategy (ES) technique to minimize the segmentation error on the testing dataset. The proposed segmentation method with the optimal parameters is tested on the experiment dataset and algorithm segmentations are compared with the ground truth. The experiment results and analysis show that algorithm segmentations are closed to the ground truth. This thesis mainly consists of the following parts: Image segmentation method based on level-set: theory and implementation. Research methodology: 3D US carotid image acquisition, ground truth, segmentation algorithm and optimization of the segmentation parameters. Experiments and data analysis. Key w