Research On The Medical Image Segmentation Based On GPU Hardware Acceleration

Medical image segmentation, especially 3D volume segmentation, is a multi-disciplinary subject and an important application of digital image processing in biomedicine engineering that relates to image and signal processing as well as medical imaging and clinical diagnosis. Meanwhile, the visualization of the volume and its segmentation result depends on the technique of computer graphics. Segmentation and visualization of medical images are widely used in diagnosis, surgery planning and simulating, radiotherapy planning and teaching in anatomy. Thus, study and research on the medical image segmentation have important significance on science and worthiness in practical application. However, since medical images are usually quite large in size and the corresponding algorithms often require lots of computation, the application area is limited by the long execution time.With the advance in computer hardware, the computation power of graphics processing unit (GPU) is growing exponentially in the latest several years. Nowadays, not only is GPU used in graphics rendering, but also in some general purpose computing beyond graphics by means of its excellent float-point arithmetic capacity, streaming parallel architecture and flexible programming. These have led to the so-called GPGPU technology.This thesis deals with the problems in medical image processing and segmentation with the help of GPU and its computation power so as to improve the performance of the program and users'experience. The thesis can be divided into two parts, concerning the medical image segmentation related application of GPU computing in graphics and general purpose respectively. In the graphics part, the thesis gives an improvement to the exising Ray Casting volume rendering implementation to make it support rendering multiple segmented objects concurrently, and also designs and implements the interface based on Ray Casting rendering for 3D volume segmentation. In the part of general purpose computing, the thesis firstly introduces the basic idea of implementing GPGPU computing through graphics rendering pipleline. Then several GPU based 2D or 3D image processing methods are implemented as examples and GPU implementations of both diffusion based and Graph Cuts segmentation are deeply studied at the ending of the thesis. The thesis also gives a comparison between GPU and CPU in the performance of several segmentation algorithm implementations by experiments, which indicates that GPU can provide great acceleration for 3D medical image segmentation.