Transfer Function Design For The Volume Visualization Of MRA Blood Vessels

With the improvement of living standard, cardiovascular disease has become one of the greatest threats to people’s lives, such as atherosclerosis, cerebral hemorrhage. Its morbidity and mortality has exceeded tumor and become the highest in the world, and it also the first cause of death in our country. Therefore, prevention and treatment of cardiovascular disease has been a very important and urgent subject. As the booming development of modern society, three-dimensional(3D) visualization technology has been widely used in the industrial process. In the field of clinical medicine, 3D medical visualization is useful, and visualization of blood vessels is significantly helpful for medical diagnosis of vascular diseases. Under the above background, this study mainly uses the volume rendering technology in medical visualization to display the vascular structures from MRA data. The main work of this dissertation can be summarized as the following three aspects:(1) Proposal of a novel transfer function according to the analysis of reqirements of designing transfer function in volume rendering. This study analyzes the deficiencies of existing transfer function, combines some algorithms about image enhancement, and proposes a novel 2D transfer function for 3D vascular display. Transfer function design has always been a difficult problem in volume rendering study. High dimensional transfer function can extract target data, but the drawback is that it is difficulty to adjust for targeted results because of too many parameters. Hence, we propose a 2D transfer function based on gray-multiscale filter. Firstly, Frangi enhancement function is used to enhance the vascular structures, making each pixel point have its corresponding scalar value, and the multi-scale filter is also constructed according to the scaling factor in Gaussian function. Secondly, considering the importance of gray level in classification of MRA data, we design the gray-multi-scale filter in 2D feature space. Finally, the characteristic regions are selected and voxels are given color and opacity, making the vascular structures displayed.(2) Further improvement of the display result by the designed 2D transfer function. Further study found that the results of MRA blood vessels visualization always have two disadvantages: one is too many fragments in visual results, and the other is that some vessels like structures block the real ones. For the first disadvantage, we used the graph connectivity calculation. According to the experimental results of different connectivity calcultation methods, we choose the 26 neighborhood method because it can eliminate the fragments well. For the second problem, we classify and sort the obtained connected sets, and give those sets different color and opacity. In this way, the vessel like structures can be removed by interaction.(3) Implementation of an accelerating computational method to solve the time-consuming problem of the proposed method. Firstly, this study conduct the ray-casting-based volume rendering acceleration based on GPU. In the practical application, doctors need to interoperate, such as rotation and scale, to get a satisfied display. But this process needs a lot of calculations, which would cause bad effects on real-time display. This paper tilizes GPU acceleration and gets the real-time rendering results. Secondly, for each voxel in our 3D data, the eigenvalues and eigenvectors of Hessian matrixes in different scales have to be calculated when constructing multiscale filter. The calcutaion process is complex, expecially for big data, and this would bring terrible inconvenience for clinical application. For this problem, we used two accelerating methods, i.e., fast computation and the algorithm improvement by Rouse-Hurwitz theorem, which shorten the calculation time significantly.