GPU-based Region Growing Segmentation Algorithm

Medical image segmentation is a popular topic. The research and development work in this area has many years of history. A large number of segmentation algorithms have been published by many scholars and most of the algorithms are accompanied with large scale datasets processing. For 2D, small scale data, it can get better result. However, with the development of various types of medical equipment and the accuracy of CT/MR, the data scale became larger and larger. For large scale 2D data and 3D data, traditional segmentation algorithm may cost a lot of time. So there are still some technical difficulties need to be resolved for real-time medical image segmentation.Currently, with the rapid development of graphics card, researches on GPU-based medical image segmentation algorithms were becoming a popular topic. Many scholars began their research and development work in this area. With the in-depth study, the topic refers to deeper area. The segmentation algorithms still have some problems need to be resolved. For example, as the restrictions of the graphics card, some segmentation algorithms which need complex data structure like region growing, are hard to be implemented in the graphics card.The goal of GPU-based medical image segmentation is to accelerate the speed of traditional segmentation algorithms by using the graphic card which is suitable for large scale parallel computing. In the clinical diagnosis with a growing number of cases, fast segmentation is a must. Although we have done many researches, there are still some difficulties need to be overcome.The goal of my master thesis is to improve and accelerate the original segmentation algorithm based on existing success.The main contributions are described as follows:As the lower speed of the existing GPU-based region growing algorithms, this paper gives out a novel index-table based segmentation algorithm to accelerate the segmentation.For the branch-shaped organ, the paper uses a segmentation algorithm by skeleton reconstruction. We first extract the skeleton of the medical image, and then reconstruct it using the skeleton. It can speed up by the GPU.