TY - GEN
T1 - An enhanced image reconstruction tool for computed tomography on GPUS
AU - Yu, Xiaodong
AU - Wang, Hao
AU - Feng, Wu Chun
AU - Gong, Hao
AU - Cao, Guohua
N1 - Publisher Copyright:
© 2017 ACM.
PY - 2017/5/15
Y1 - 2017/5/15
N2 - The algebraic reconstruction technique (ART) is an iterative algorithm for CT (i.e., computed tomography) image reconstruction that delivers better image quality with less radiation dosage than the industry-standard filtered back projection (FBP). However, the high computational cost of ART requires researchers to turn to highperformance computing to accelerate the algorithm. Alas, existing approaches for ART suffer from in efficient design of compressed data structures and computational kernels on GPUS. Thus, this paper presents our enhanced CUDA-based CT image reconstruction tool based on the algebraic reconstruction technique (ART) or cuART. It delivers a compression and parallelization solution for ART-based image reconstruction on GPUS. We address the under-performing, but popular, GPU libraries, e.g., cuSPARSE, BRC, and CSR5, on the ART algorithm and propose a symmetrybased CSR format (SCSR) to further compress the CSR data structure and optimize data access for both SpMV and SpMV T via a column-indices permutation. We also propose sorting-based and sorting-free blocking techniques to optimize the kernel computation by leveraging the sparsity patterns of the system matrix. The end result is that cuART can reduce the memory footprint significantly and enable practical CT datasets to fit into a single GPU. The experimental results on a NVIDIA Tesla K80 GPU illustrate that our approach can achieve up to 6.8x, 7.2x, and 5.4x speedups over counterparts that use cuSPARSE, BRC, and CSR5, respectively.
AB - The algebraic reconstruction technique (ART) is an iterative algorithm for CT (i.e., computed tomography) image reconstruction that delivers better image quality with less radiation dosage than the industry-standard filtered back projection (FBP). However, the high computational cost of ART requires researchers to turn to highperformance computing to accelerate the algorithm. Alas, existing approaches for ART suffer from in efficient design of compressed data structures and computational kernels on GPUS. Thus, this paper presents our enhanced CUDA-based CT image reconstruction tool based on the algebraic reconstruction technique (ART) or cuART. It delivers a compression and parallelization solution for ART-based image reconstruction on GPUS. We address the under-performing, but popular, GPU libraries, e.g., cuSPARSE, BRC, and CSR5, on the ART algorithm and propose a symmetrybased CSR format (SCSR) to further compress the CSR data structure and optimize data access for both SpMV and SpMV T via a column-indices permutation. We also propose sorting-based and sorting-free blocking techniques to optimize the kernel computation by leveraging the sparsity patterns of the system matrix. The end result is that cuART can reduce the memory footprint significantly and enable practical CT datasets to fit into a single GPU. The experimental results on a NVIDIA Tesla K80 GPU illustrate that our approach can achieve up to 6.8x, 7.2x, and 5.4x speedups over counterparts that use cuSPARSE, BRC, and CSR5, respectively.
KW - Algebraic reconstruction technique
KW - Computed tomography
KW - GPU
KW - Image reconstruction
KW - SpMV
KW - Sparse matrix-vector multiplication
KW - Transposition
UR - http://www.scopus.com/inward/record.url?scp=85027054141&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85027054141&partnerID=8YFLogxK
U2 - 10.1145/3075564.3078889
DO - 10.1145/3075564.3078889
M3 - Conference contribution
AN - SCOPUS:85027054141
T3 - ACM International Conference on Computing Frontiers 2017, CF 2017
SP - 97
EP - 106
BT - ACM International Conference on Computing Frontiers 2017, CF 2017
T2 - 14th ACM International Conference on Computing Frontiers, CF 2017
Y2 - 15 May 2017 through 17 May 2017
ER -