Simulation of X-ray Attenuation on the GPU

摘要: Abstract In this paper, we propose to take advantage of computer graphics hardware achievean accelerated simulation X-ray transmission imaging, and compare results with afast robust software-only implementation. The running times the GPU CPUimplementations are compared in different test cases. show that GPUimplementation full floating point precision is faster by a factor about 60 65than CPU implementation, without any significant loss accuracy. increase inperformance achieved calculations opens up new perspectives. Notably, it pavesthewayforphysically-realisticsimulationofX-rayimagingininteractivetime. Categories Subject Descriptors (accordingtoACMCCS):I.3.5ComputerGraphics:Physically based modeling; I.3.7 Computer Graphics: Raytracing; J.2 Applica-tions: Physics. Keywords: Three-Dimensional Graphics Realism, Raytracing, Physical Sciences andEngineering,Physics. 1 Introduction imaging techniques such as radiography or tomography extensivelystudied physics community physically-based codes available.Deterministic methods on ray-tracing commonly used compute direct images (i.e.images formed beam transmitted interaction through scanned object)ofcomputer-aideddesign(CAD)models. Ray-tracingprovidesafastalternativetoMonteCarlomethods [4]. Such programs very useful optimize experiment parameters, conceiveimagingsystems,ortotakeintoaccountnon-destructivetestingduringthedesignofamechanicalstructure[1,10]. However,evenwithfastraytracingalgorithms,thesimulationofcomplexX-rayimagingsystemsstillrequiresverylongcomputationtimesandisnotsuitableforaninteractiveuseaswouldberequiredinamedicaltrainingtool.Physics-basedsimulationsaretraditionallyperformedonCPUs. However,thereisagrowinginterestforgeneral-purposecomputationonGPUs(GPGPU)andthishasbeenanactiveareaofresearchsometime[13].Inthispaper,wepresentanefficientsimulationofX-rayattenuationthroughcomplexobjects,thatmakesuseofthecapabilityimprovementoftoday’sgraphicscards. Wealsocomparetheper-formanceofthisGPUapproachwithanefficientsoftware-onlyimplementation. ToourknowledgethisisthefirstGPU-basedX-Rayattenuationsimulation. Suchasimulationtoolcanbedeployedinmedicalvirtualinteractiveapplicationsfortrainingfluoroscopyguidanceofneedles,cathetersand guidewires [18], can also be speed-up current physics-based wherecomputationalaccuracyiscritical.