基于压缩感知CS（ Compressed Sensing ）理论的稀疏磁共振图像MRI（ Magnetic Resonance Imaging ）重构算法包含大量的浮点运算，重构所花费的时间要远远大于傅里叶正反变换重构算法。针对该问题，利用图形处理器GPU（ Graphic Processing Unit ）强大的并行处理能力，在NVIDIA CUDA（ Compute Unified Device Architecture ）的框架上对正交匹配追踪OMP（ Orthogonal Matching Pursuit ）算法进行并行化的设计与实现。实验结果表明，基于GPU实现的算法具有较高的迭代重构速度，对10242大小的磁共振图像的重构仅为1．4秒，是CPU实现的24倍，可以满足实际应用对实时性的要求。%Sparse MRI image reconstruction algorithm based on compressed sensing theory contains a large number of floating point operation, and is much time-consuming than the traditional inverse Fourier reconstruction algorithm in reconstruction process .To solve this problem, we make use of the powerful parallel processing capability of graphic processing unit ( GPU) to carry out the parallel design and implementation on orthogonal matching pursuit ( OMP ) algorithm based on the framework of NVIDIA computer unified device architecture ( CUDA ) . Experimental results show that , the algorithm implemented based on GPU has higher iterative reconstruction speed , the reconstruction of an MRI image with 1 0242 size only takes 1.4 s, 24 times faster than the CPU implementation , this can meet the demand of practical application in real-time property .
展开▼
