A Metropolis Monte Carlo Simulation Scheme For Fast Photon Transport In Cone Beam CT

Cone-beam computed tomography(CBCT)is one of the most advanced and hottest technology in modern medical CT imaging filed.Nevertheless,the problem of photon scattering seriously restricts the clinical application and development of CBCT images.It is commonly known that two software correction methods are introduced to estimate photon scattering distribution,and thus the problem is solved.One of the methods is deconvolution method using scattering kernel,and the other one is conventional Monte Carlo(MC)method.Unfortunately,scattering kernel is hard to design in deconvolution method,resulting in inaccurate scattering distribution.Although conventional MC method does well in scattering distribution estimation,it suffers from low efficiency and slow convergence and cannot be used in clinical application.Therefore,how to realize fast scattering distribution estimation becomes hot spot at present in the field of medical physics.In this paper,according to the analysis and reviews of physics of photon scattering and conventional MC simulation method,it is found that conventional MC simulations are simulated using a passive independent sampling strategy in photon scattering.Under such a strategy,conventional MC simulations unavoidably spends plenty of time in transporting those particles that do not contribute to the signal of interest at the detector,yielding a low computational efficiency and hindering its clinical application.Focusing on the bottleneck problem of existing MC methods,a novel and active MC method,named gpu-based Metropolis Monte Carlo(gMMC),is proposed to promote photon scattering simulation efficiency with different sampling principle in this paper.The proposed method positively takes simulation tasks into the controllable photon path variation space,and samples the photon path by using the automatic sampling model.The Metropolis path sampling algorithm is utilized to compare the "importance" of sampling paths before and after.The relatively important photon path is then selected as the energy deposition path.Therefore,the auto importance sampling strategy is achieved.This controllable path sampling method regard the entire photon path as simulation object,which is different from the independent sampling strategy.To further improve the efficiency,GPU parallel computing architecture is utilized on hardware side.As a result,simulation cases and clinical cases indicate that the proposed method greatly improve photon scattering simulation efficiency.In detail,the scatter simulation was accelerated 20?48 times compared to conventional MC scheme with relative difference less than 3%.Without utilizing de-noise algorithm,the simulation time is decreased to mins level.Aiming to solve the problem of low efficiency of conventional MC method,a novel MC simulation algorithm is developed in this paper and is proved to improve the efficiency of photon scattering simulation significantly.Although some preliminary achievements have been made in this paper,it still needs to be further explored and perfected for the actual clinical application.