Research On Medical Image Volume Rendering And Its Acceleration Technique

With the improvement of visualization technology, medical image visualization is becoming more and more useful in the aspect of disease diagnosis,surgical navigation and computer-aided education. Direct volume rendering renders the volume data set directly, which is able to display the inner hierarchical relations of the object and to exhibit plentiful 3D information. So, volume rendering still has more room for development and good development prospects.Based on the systematic research of basic principle and classical rendering methods of Direct Volume Rendering, the critical aspects are described in details. Suggestions are brought out to solve the problems in these aspects, which can meet the need of interactive rendering.The design of transfer function plays an import role in volume visualization that will highlight those parts of the medical data which are of interest to the user. One-dimensional transfer function is difficult to extract the complex feature, as a result, a multidimensional transfer function based on scalar value and the gradient magnitude appears, which is on the basis of boundary model. The boundary between objects appears in the form of arc in the histogram. If two arc overlapping, it would be difficult to select arc. So a new histogram, which is called LH histogram, is introduced to volume rendering. In the LH histogram, arc becomes blob, and transfer function based on LH histogram solves the problem that is difficult to select overlapping arc in the previous histogram. However, the original method to compute the LH value could not meet the need of interaction rendering. What's more, it is easy to lose some important boundary, if only through the gradient magnitude criterion. Integral path is susceptible to noise, which leads to decreased robustness.This paper brings out a new calculation method, which is faster than the original one. Then it weights by the distance of voxel to object boundary, which improves the accuracy and robustness of the LH histogram.Volume rendering acceleration technique is always the key spot. This paper outlines the multi-pass and single-pass ray casting algorithm based on GPU. Although its performance greatly exceeds the CPU execution, it often has to abandon the flexibility of programming when in order to achieve the ideal rendering speed, that's because the defects of its structure. The Common Unified Device Architecture (CUDA) is a fundamentally new programming approach making use of the unified shader design of most current GPU.In this work the new technology is applied to volume ray casting algorithm in the field of medical image processing. The normal GPU-based ray casting algorithm is improved and rendering process is redesigned into Single Instruction Multiple Data (SIMD) mode. Finally suggestions are presented on how to optimize the algorithm. The experiment results show that it can achieve high quality image and interactive frame rates by using CUDA.This paper also analyzes the diffuse and specular reflection in details in the common illumination model, and comparison of three kinds of gradient calculation method, which can affect lightshade.