Medical Image Volume Rendering Research Based On GPU

The calculated or measured3D data can be rendered on the screen by the technology ofdirect volume rendering, which provides the most reliable and useful tool for researchers toobserve the data. Volume rendering technology is widely applied and fast developed inmedical image processing. In recent years, with the rapid development of modern graphicsprocessor (GPU), the ability of processing of GPU is more powerful than that of CPU, whichmakes it possible for real-time volume rendering. Volume rendering based on GPU hasbecame a research hotspot, however, current popular technology of volume rendering still cannot effectively achieve results due to the complexity and emerging dynamics of the medicaldata.In addition, it is also a worth studying problem that how to build a friendly user interface,allow the operator to obtain the3D reconstruction image by the volume renderingvisualization system more easily and more flexibly.Intended for medical images application requirements, this article focus on solvingproblems like, fast real-time display, effective display information the researchers interested,fast real-time dynamic volume rendering, and convenient visualization system interface. Inthis paper, we study fast real-time high-quality volume rendering based on GPU technology,and explore acceleration technology, high dimensional transfer function design, as well as the4D dynamic display and3D visualization system design in depth.The main research results are obtained as follows in this paper:1) The real-time ray casting volume rendering algorithm based on GPU is discussed andimplemented in this paper, the principle of volume rendering and render process arestudied, the system with flexible volume rendering framework is implemented by theCg and OpenGL,kinds of rendering result such as general drawing, maximum densityprojection drawing, isosurface rendering, Phong illumination rendering and othercommonly used drawing effect are implemented. And further the real-timeacceleration method is studied, containing method of light early termination and clipinvalid voxels method based on octree is further studied, experimental results showthat using the accelerated algorithm does not affect the rendering effect, but increaseon the basis of the volume rendering speed greatly.2) To render the region of interest better, it is necessary to design a suitable transferfunction. Interaction design based on high dimensional histograms of highdimensional transfer function method is a popular method at present, but the methodis complex and the effect is not satisfied. To solve the problems of high dimension transfer function design, a method of volume rendering of automatic and interactivedesign of high dimensional transfer function based on K-Means++clusteringalgorithm is presented in this paper. Experimental results show that this method caneliminate the complex of the design of high dimension transfer function, and varioushuman tissues structure characteristics can be showed in the rendering results.3) The traditional3D medical volume rendering technology cannot satisfy the needs ofdynamic4D medical volume rendering, because the large amount of4D medicalimage data make it difficult to achieve high performance real-time rendering bytraditional technology. Focus on the demand, the method of dynamic4D fast volumerendering based on GPU is presented and implemented in this paper, the4D dynamicdrawing process is also presented. The performance of the proposed algorithm is agreater degree of improvement versus the traditional light projection algorithmwithout declining volume rendering quality.4)3D visualization system based on GPU is finally constructed based on above studies.The main views and auxiliary views, operations such as related rotation, scaling,cutting, as well as the interface interaction design of the transfer function areimplemented. In order to obtain highly flexible interactivity and real-timeperformance in cutting operation, an improved cutting algorithm based on thevolume mark is presented. GPU assistant resources are used in the algorithm. Thealgorithm does not have to change volume rendering process, drawing,and has thevery high interactivity at the same time.The fast real-time high quality volume rendering based on GPU technology is discussedin this paper, and the acceleration technology, higher dimensional transfer function design, thefour-dimensional dynamic display and3D visualization system design are explored deeply inthis paper. The work stated in this thesis has good meanings for medical practice and providesa good reference and new ideas for researcher.