Meeting Banner
Abstract #2266

Fast Temporally Constrained Reconstruction on a GPU Cluster

Jordan P. Hulet1, 2, Ganesh Adluru, Julio C. Facelli1, Edward DiBella, Dennis L. Parker, 3

1Biomedical Informatics, University of Utah, Salt Lake City, UT, United States; 2Utah Center for Advanced Imaging Research, Salt Lake City, UT, United States; 3Utah Center for Advanced Imaging Research, University of Utah, Salt Lake City, UT, United States

In this work, the reconstruction time required for temporally constrained reconstruction of a large cardiac perfusion data set was reduced by utilizing 12 graphics processing units distributed across 6 different compute nodes. A speedup of over 30x was achieved compared to a 16 core CPU system resulting in a reconstruction time of only 48 seconds.

Keywords

accelerate achieved acquisition administration advanced allow allowed allowing among approval architecture assessed attempted available boundaries card cardiac care city clinical code coil coils collected complex component components compressed computational computationally compute computing configuration constrained constraint converge core cores cost count created dataset deal definitions demand demanding determined dimensionality distribute distributed divided drastically drawback estimation exam fast foundation frames gradient grant graphics health hours hypothesized implemented impractical included individual informatics interface iris iterations lake makes making mask measured message minute minutes necessary need node nodes noted operations optimal overlap overlapping parallel parallelized parker part passing patient perfusion pixel preprocessing principal prior processing quality radial reasonable receiver reconstructed reconstruction reconstructions recovery reduce reduced reduction repeat repetition repetitions required restrictions review salt sampled sampling saturation scenarios sensing sensitivities sensitivity sets several shorter significantly similarly simultaneously speed speedup splitting still sufficient supported system systems take temporal temporally thread threaded threads training units utility views virtual