基于GPU加速的并行脑皮层重建算法研究  

Parallel algorithm for cortical reconstruction based on GPU

作　　者：连远锋[1] 赵琰[1] 何晖光[2] 吴发林[1] 

机构地区：[1]北京航空航天大学仪器科学与光电工程学院,北京100191 [2]中国科学院自动化研究所,北京100190

出　　处：《仪器仪表学报》2013年第4期866-872,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然基金(60701021);中国石油大学(北京)基础科研基金(JCXK-2010-02)资助项目

摘　　要：提出了一种基于GPU加速的脑皮层重建并行算法。该算法采用过完备球面小波变换对大脑皮层数据进行多尺度分解及重建。首先在三维欧氏空间,给出了沿坐标轴独立进行球面调和-卷积计算的理论证明;然后利用GPU的通用计算功能对球面调和变换和球面卷积运算进行并行处理,实现了脑皮层快速重建。实验结果表明,使用提出的算法重建的脑皮层与传统CPU算法的结果几乎没有差别,但是计算速度提高了50倍。This paper presents a fast GPU-based cortical reconstruction algorithm. The approach uses multi-resolution Over-Complete Spherical Wavelets （OCSW） decomposition and reconstruction for human cortical surface. Based on the Euclidean three-space, this paper first give proof which ensures the correctness of executing spherical harmonic and spherical convolution by independently scanning each axis. Then the proposed algorithm is implemented in paral- lel on GPU to achieve high performance of cortical reconstruction. Experimental results show that the cortical surfaces generated with the proposed GPU algorithm are indistinguishable with those generated with traditional CPU algo- rithms, while the speed of the proposed method is at least 50 times faster than that of the traditional CPU algorithm.

关 键 词：图形处理器 通用计算 过完备球面小波 球面调和 脑皮层重建 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]