The Research And Implementation Of Medical Image Three-dimensional Reconstruction Algorithm Based On GPU

Along with the constantly development of scientific computing visualization technology, the medical image three-dimensional reconstruction technique begin to play a very important role in the field of disease diagnosis, surgical navigation and medical education. Especially volume rendering algorithm can show the inner hierarchy and structural characteristics of the tissues and organs. By showing the rich information of three-dimensional reconstruction and a strong feeling of realistic, volume rendering algorithm has a great space and good prospect to develop.Due to the large scale of medical image, the traditional three-dimensional reconstruction algorithm based on CPU can not achieve the real-time demand. In recent years, with the rapid development of GPU-based general purpose computing, making a new way to solve this problem. The ray casting algorithm as a typical medical image reconstruction method has the features to transform from CPU to GPU in order to improve efficiency. This paper is focus on this issue to study and analyze.This paper first analyzes two programming models of GPU-based programming, Cg language and CUDA architecture. Cg language makes the shader code embedded in graphic rendering pipeline to achieve the control of processor and memory in GPU chip. While the CUDA architecture is a new general purpose parallel computing programming model to achieve a better control of processor and memory in GPU chip. Based on accelerate technology of volume rendering algorithm. This paper analyzes the accelerate principle of ray casting algorithm. And introduce two accelerated ray casting algorithm based on GPU, they are multi-pass ray casting algorithm and single-pass ray casting algorithm. Based on studying these two methods, this paper implements a new single-pass ray casting algorithm. It gets the max path value of the projected light line by using depth value of texture cube and FBO object. Experiment result show that this method can improve efficiency of ray casting algorithm.This paper analyzes the CUDA architecture, design idea and programming model and according to its characteristics to improve the tradition accelerated ray casting algorithm. Firstly this paper analyzes key issues of how to implement ray casting algorithm on CUDA architecture, and then according to the feature of experiment platform hardware environment to design CUDA threads allocation scheme, finally by using PBO object and shared memory to design a accelerated ray casting algorithm based on CUDA architecture. Experiment results show that this method can complete medical image three-dimensional reconstruction very fast in high quality form.This paper also analyzes and compares the experiment results of two GPU based accelerated algorithm to conclude that whether in reconstruction effect or rendering efficiency, the CUDA architecture is the best choice for medical image three-dimensional reconstruction algorithm to accelerate.