Research On 3D Real-time Rendering Technology Of Brain Imagesbased On CUDA

The three-dimensional medical image rendering technology has always been a hot topic at home and abroad.Nowadays,for the huge medical imaging data,how to further improve image quality and increase the rendering speed is even more challenging.There are shortcomings in the traditional CPU-based and GPU-based such as the earlier GPGPU(General Purpose GPU)rendering algorithms.For example,the computing is incompatible with the hardware architecture,or programming for algorithm is very difficult,making these technologies not widely used in medical image rendering.Furthermore,it will take more time to diagnose brain diseases and observe the brain tissue image by using traditional medical image 3D rendering methods.For the structure of human,brain is relatively more complex than other organizational structures,which brings a lot of inconvenience to the medical and scientific research.In this dissertation,current common three-dimensional medical image rendering algorithms were compared and analysed.According to the human brain structure and the basic knowledge of parallel acceleration technique based on CUDA(Compute Unified Device Architecture),the volume rendering algorithm was optimized to acquire a better performance.And,the experimental results showed that the improved algorithm based on the CUDA architecture significantly improved the three-dimensional rendering speed.The main contents are as follows: firstly,the original brain image data was obtained from Alzheimer's disease neuroimaging database,and Voxel-Based Morphometry methods and Fuzzy C-Means clustering algorithm were used to segment the brain magnetic resonance image into three parts,including gray matter,white matter,and cerebrospinal fluid.The segmentation results using different methods were compared and analysed for subsequent 3D reconstructions.Secondly,The Ray Casting algorithm was used for 3D image rendering,while the rendering quality and the processing time were observed.Finally,the key technology of parallel image rendering was used to achieve parallel processing on a CUDA system and the algorithm was optimized mainly in the following several aspects: memory access,instruction stream,thread allocation,and data transmission,in order to get greater memory bandwidth,greater instruction throughput and shorter computational time.The 3D model of the brain tissue image using parallel algorithm based on CUDA was also rendered.The time wasdocumented and was used to compare with that of without the use of CUDA parallel computing architecture.It is shown by the recorded rendering time that using the parallel algorithm on the CUDA platform to render three-dimensional model not only can guarantee the quality,but also can improve the three-dimensional rendering speed by dozens of times,which has important significance to the diagnosis and scientific research.