Abstract

Info

Title:: GPU acceleration for kernel regression algorithm of three-dimensional medical image

Author(s):: WANG Yukun1; LIU Rong1; 2; WEN Tiexiang2; LI Ling2; 1. School of Computer Science and Technology, Henan Polytechnic University, Jiaozuo 454000, China; 2. Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China

Keywords:: ; Keywords: GPU acceleration; CUDA programming; three-dimensional classical kernel regression; three-dimensional steering kernel regression

Abstract:: Abstract: Kernel regression which has been widely used in the field of medical image processing and medical image reconstruction, achieving many significant results, includes classic kernel regression (CKR) and steering kernel regression (SKR). Three-dimensional (3D) SKR has better denoising effects and edge-preserving results than 3D CKR. At present, GPU-based 3D CKR algorithm has been used on medical image reconstruction, while the computational complexity of 3D SKR algorithm limits the application of 3D SKR algorithm. The realization of GPU-based 3D CKR algorithm is a valuable research topic and remains challenging. In this experiment, the calculation process of 3D SKR algorithm is firstly optimized, and then CUDA programming is used to implement the GPU-based 3D SKR parallel acceleration algorithm. The experiments show that the speedup ratio of 3D SKR algorithm based on GPU is about 244.9-246.3 as compared with single-threaded CPU-based 3D SKR algorithm, and about 123.0-137.4 as compared with the multi-threaded CPU-based 3D SKR algorithm.