The existing Sparse MRI reconstruction algorithm is made practical by parallelizing it under the framework of NVIDIA CUDA using GPGPU(General Purpose GPU).A big problem preventing compressed sensing based Sparse MRI from being applied to practice is the reconstruction time due to massive floating-point operation. A speedup up to 76 times on NVIDIA GTX275 GPU against the Intel Q8200 CPU is gained to reduce the processing time from more than 4 minutes to about 3.4 seconds.%利用GPGPU(General Purpose GPU)強(qiáng)大的并行處理能力,基于NVIDIA CUDA框架對(duì)已有的稀疏磁共振(Sparse MRI)重建算法進(jìn)行了并行化改造,使其能夠適應(yīng)實(shí)際應(yīng)用的要求.稀疏磁共振成像的重建算法包含大量的浮點(diǎn)運(yùn)算,計(jì)算耗時(shí)嚴(yán)重,難以應(yīng)用于實(shí)際,必須對(duì)其進(jìn)行加速和優(yōu)化.實(shí)驗(yàn)結(jié)果顯示,NVIDIA GTX275 GPU使運(yùn)算時(shí)間從4分多鐘縮短到3.4秒左右,與Intel Q8200 CPU相比,達(dá)到了76倍的加速.
展開▼