Premium
Poster — Thur Eve — 34: Accelerated Analytical Scatter Estimation with Graphics Processing Units
Author(s) -
Ingleby H,
Lippuner J,
Elbakri I,
Rickey D
Publication year - 2010
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.3476139
Subject(s) - monte carlo method , imaging phantom , computer science , estimator , computational science , cuda , algorithm , voxel , matlab , graphics , optics , physics , computer graphics (images) , artificial intelligence , parallel computing , mathematics , statistics , operating system
Image degradation due to scatter can be a serious problem in x‐ray imaging, particularly in cone‐beam computed tomography because of the high scatter to primary ratio. Computational methods to estimate scatter are useful both for system modeling and optimization as well as algorithmic scatter correction. Computational scatter estimators are generally based on either Monte Carlo simulation or analytical calculations. Monte Carlo methods can incorporate very accurate models of interaction physics, but are typically very time consuming. Analytical methods, while usually less accurate than Monte Carlo due to simplifications required to render them computationally tractable, are more amenable to acceleration by parallelization. We previously developed an analytical method for estimating Compton and Rayleigh single scatter for a voxelized phantom in a cone beam geometry. Scatter estimates produced with our initial Matlab code showed good agreement with those obtained from Monte Carlo simulation of an identical imaging geometry in EGSnrc. Computation time with the analytical code was still significant, however, especially when using a high‐resolution phantom with small voxels. Our goal for this project was to accelerate our analytical scatter estimator, without loss of accuracy, by porting the code for use with Nvidia graphics processing units (GPUs) with the CUDA programming environment. Using four GPUs, we obtained speed‐up factors of approximately 700X relative to the original Matlab code running on a single CPU while maintaining good agreement with our reference Monte Carlo results. We plan to apply our GPU‐based analytical scatter estimator to a scatter correction algorithm for cone beam computed tomography.