Research On GPU-based Medical Image Real-time Volume Rendering Technology

Visualization in scientific computing was proposed in1980s.The visualizationapplication of three-dimensional data is becoming more and more widespread, and ithas been developed to the area of volume rendering. Volume rendering is an importantdirection of scientific visualization. Volume rendering is regularly used in the field ofmedical image processing. The study on volume rendering of medical images hasincreased year by year. Because of the development of computer technology and theimprovement of medical imaging equipment, the slice spacing after scan is becomingthinner and thinner, and the resolution is higher. So the data size is larger. Especially,with the appearance of4DCT, the data it scans each time is up to10times or more asbig as three-dimensional data. The traditional rendering technology, which appliesCPU to performing all of the compute, can’t satisfy demands of real-time renderingand interaction. In recent ten years, GPU has acquired considerable development andhas been moving forward to the direction of general purpose computing. The dataprocessing property of GPU far exceeds that of CPU, which makes the realization ofvolume rendering in general computer possible. At present, though the study ofthree-dimensional reconstruction algorithms is much more, the transfer function of alarge majority of algorithms renders special effect with special datasets. Fewalgorithms can render different effects according to different datasets. In addition, dueto without time-varying information, three-dimensional reconstruction can’t showfocus varying with the time. Doctors can only make a comparison of collected dataduring different time by hand, which wastes time and energy. So how to performfour-dimensional visualization with data collected from different time is of greatimportance. However, domestic studies aimed at four-dimensional visualization arestill in its beginning stage. When four-dimensional visualization is performed ingeneral computers using the existing algorithms, it can’t be rendered real-timely andinteracted fluently. Thus, as to the two questions proposed above, the paper conductsrelevant researches. The traditional algorithms have the following shortages in realizing medicalimage volume rendering, such as long time consuming, without real-time interactionand single rendering effect. In order to overcome these shortages, a ray castingalgorithm based on GPU was realized in the paper and it is used to the real-timevolume rendering of medical images. The algorithm studies the distribution of humanCT value and designs different color and opacity transfer functions. So it can switchesamong rendering effects of different organs. The experiment result shows when thenumber of slices is up to900, the rendering time of the algorithm proposed in thepaper can be controlled to2seconds or so, the interaction speed can be up to morethan20fps. So it absolutely satisfies demands of clinical diagnosis and treatment.Moreover, the paper made comparisons of rendering time and quality on ray castingalgorithm based on CPU and GPU. The acceleration ratio can be up to9times at mostand the rendering quality based on GPU is higher.With medical images, the general computer can’t real-timely performfour-dimensional visualization and fluently interact. To solve the two problems, thepaper proposed a solution, which is four-dimensional visualization of medical imagesand extraction of focuses based on GPU. With the assistant of OpenGL, the paperrealized the read and display of volume data by programming. According to thedifferent grey level of different organs, the bound of grey level was set in the colorlookup table to control the display of different organs. In the condition of determiningthe position of focuses, the extraction of focuses and local four-dimensionalvisualization were realized. As to the10time phases data of lung, the rendering speedof the algorithm can be up to20fps or so. The rendering speed of focuses afterextraction can be up to31fps. It can be maintained at20fps or so even in the processof interaction, which realizes fluent play and completely satisfies the clinical demandsof doctors.