Three-dimensional Medical Image Segmentation And Vascular Contrast Enhancement Surgery Study

Computed tomography ( CT ) is a noninvasive technique for producing cross-sectional images of a subject, and is now widely used for medical applications. Today's CT scanners can produce hundreds of images in a single study, so computer-aided analyzing and visualization techniques draw more attention than before.Image segmentation and vessel enhancement are two important problems in the application of medical images. Among many segmentation algorithms of medical images, which can be categorized to edge-based methods and region-based methods, no universal method exists, either does an evaluation criteria. Blood vessel visualization of medical images plays an important role in diagnosis, but because of noise and other reasons, vessel enhancement procedures should be followed to improve the visualization result.This thesis will focus on the CT images of liver—one of the most important organ of human body. Aiming to solve these problems, the work of the thesis is shown below:1. The 3D MRF-EM algorithm based on Markov Random Field theory is proposed to segment the original CT images. This algorithm is region-based, and boasts computing the whole 3D data at a time and utilizing spatial information of data.2. Mathematical Morphology and Region Growing are used to refine the rough segmentation results of MRF-EM algorithm. This refinement segmentation can fill holes in liver, smooth the surface, and remove non-liver parts.3. Three algorithms are combined with Maximum Intensity Projection (MIP) algorithm to improve vascular structure of liver. The first method is Mixtured Gaussian filtering, which is a mixture of two lowpass filters and can greatly reduce the noise in the image data. The second method is Local Maximum Mean, which utilizes the directional properties of vascular structure and can enhance the small and disjointed vessels of liver. The third method is using eigenvalues of Hessian matrix, which utilizes relationship between eigenvalues and tubular structure to extract vascular structure from 3D data. Then vascular structure model is set up to do Performance Analysis, which can compare and evaluate these three vessel-enhancing algorithms. The developed algorithms were tested using a 89-slices dataset from American GE Medical Systems. The results demonstrate that the proposed procedures can effectively segment 3D CT images and enhance vascular structure of liver.