Fast 3D Ultrasonic Imaging Research Based On CUDA

In the field of medical image, three-dimensional ultrasound imaging can visually display the characteristics of the objects, facilitating medical diagnosis. However, because volume data is to be processed in the three-dimensional ultrasound imaging, the amount of calculation and data are much larger than that of the two-dimensional ultrasound imaging. The slow speed of conventional imaging methods seriously impedes the development of three-dimensional ultrasound imaging system. Nowadays, with the development of computer technique, CUDA technology provides a new effective way to increase the speed of three-dimensional ultrasound imaging. CUDA technology is a parallel computing architecture developed by NVIDIA, which features the use of GPU parallel execution of multiple threads to achieve large-scale computing. After analyzing the CUDA accelerated method, this paper applies the CUDA acceleration technology to three-dimensional ultrasound imaging, which increases the speed of the imaging and cuts the cost of the system. Our work is organized as follows:1. The entire body of data can be copied from the CPU-side to the GPU-side using CUDA stream. The overlapping function of GPU device can hide the transmission time of the data, speeding up the transfer.2. Three-dimensional volume reconstruction is achieved based on GPU, and compared with the speed and quality of traditional CPU imaging method. In addition, we carry out a research on a de-noising method based on threshold segmentation and anisotropic diffusion.The threshold segmentation method preprocesses the raw data collected by ultrasound probe,and anisotropic diffusion filter is utilized image reconstruction after the body for treatment to improve the quality of the reconstructed image de-noising. The method improves the quality of volume rendering image.3. In the process of volume render, firstly, we analyze the efficiency of rendering with different CUDA memory banks storing data, then achieve a fast GPU-based ray-casting algorithm. Meanwhile, an improved variable step size method which determines the current sampling point through relationships with empty voxels and gray value threshold of the next sampling point sampling step. Without losing the edge voxels case, using a larger sampling step to further improve the rendering efficiency.4. The experiments verify the effectiveness of the accelerated method proposed in the paper in data transmission, body reconstruction and volume rendering process. In light of the analysis of experimental results, Fast three-dimensional ultrasound imaging methods described herein implemented, to ensure the image quality of the premise, greatly improve the imaging speed, basically meet the medical requirements for real-time imaging.