Chinese Journal of Lasers, Volume. 51, Issue 21, 2107104(2024)
Comprehensive Review of Acceleration Techniques for Monte Carlo Simulations of Photon Transport in Biological Tissues
Figures & Tables(8)
Fig. 1. Publication volume of MC simulation photon transport acceleration from 1993 to 2023. (a) Cumulative and annual publication volume in MC simulation photon transport acceleration field; (b) annual publication volume in different MC simulation photon transport acceleration methods
Fig. 2. Research state of MC simulation photon transport acceleration from 1993 to 2023. (a) Research status in different countries;
Fig. 4. Comparison of relative error and acceleration factor for different acceleration methods. (a) Comparison of relative error;
Fig. 5. Co-occurrence graph of keywords in MC simulation photon propagation acceleration field from 2013 to 2023
Table 1. Algorithm acceleration
View tableTable 1. Algorithm acceleration
Algorithm Author Tissue model Application and effect Baseline
fitting
method
Graaff et al.［ 46 ］Layered tissue， semi-infinite medium Photon density Kienle et al.［ 47 ］Homogeneous tissue， semi-infinite medium Diffuse reflection process， with an absorption rate error of less than 2% Pifferi et al.［ 48 ］Homogeneous tissue， semi-infinite medium Diffuse reflectance and transmittance Perturbation method Sassaroli［ 53 ］Non-uniform tissue Absorption and scattering coefficients Hayakawa et al.［ 52 ］Double-layer tissue Extract derivative information from a single MC simulation Kumar et al.［ 60 ］Non-uniform tissue Rapid construction and reconstruction Yao et al.［ 61 ］Non-uniform tissue Reproduce the photon path Hybrid
method
Zhu et al.［ 54 ］Layered tissue Wider range of detectors and tumor models Flock et al.［ 33 ］Non-uniform tissue Correction functions from diffusion theory Wang et al.［ 34 ，62 ］Layered， non-uniform tissue， finite-thickness medium Achieving seven times the speed with errors within two standard deviations Alexandrakis et al.［ 63 ］Double-layer tissue Faster than standard MC simulations by hundreds of times Hayashi et al.［ 64 ］Non-uniform tissue， head model Region-divided hybrid method in head models Donner et al.［ 65 ］Layered， non-uniform tissue Numerical relationship between each layer’s optical parameters Luo et al.［ 31 ］Non-uniform tissue Reduce errors at the surface of high-absorption media Tinet et al.［ 66 ］Non-uniform tissue Use statistical information to analyze computational results Chatigny et al.［ 67 ］Thick layered tissue Standard MC method combined with isotropic diffusion similarity criteria Wu et al.［ 68 ］Non-uniform tissue A model in X-ray medical applications Yan et al.［ 69 ］Non-uniform tissue Combine grid and voxel domain information while being
2‒6 times faster
Variance reduction
method
Liu et al.［ 56 ］Double-layer tissue Use geometric segmentation Chen et al.［ 70 -71 ］Thick homogeneous tissue Increase the sampling of photons likely to intersect the detector Behin-Ain et al.［ 72 ］Thick scattering medium Increase the speed of the temporal point spread function by at least four times Lima et al.［ 58 -59 ］Non-uniform tissue Improving importance sampling method， reducing computation time by up to three orders Golosio et al.［ 73 ］Non-uniform tissue Reduce computation time by several orders of magnitude in X-ray fluorescence spectroscopy Williamson［ 74 ］Non-uniform tissue Addressing issues like photon cross-section data selection
Table 2. Parallel computing acceleration
View tableTable 2. Parallel computing acceleration
Parallel unit Author Tissue model Application and effect CPU Colasanti et al.［75］ Homogeneous tissue， semi-infinite medium Run on computers with multiple processors GPU Alerstam et al.［76］ Semi-infinite medium Achieve speed 1000 times faster Alerstam et al.［77］ Semi-infinite medium Overcome performance bottlenecks of atomic access Martinsen et al.［80］ Inhomogeneous medium MC simulation on an NVIDIA® 8800GT graphics card， 70 times faster Fang et al.［81］ Arbitrary 3D inhomogeneous medium Achieve speed 300 times faster than traditional CPU methods using 1792 parallel threads Ren et al.［28］ Triangular mesh model，non-uniform tissue Algorithm for complex heterogeneous tissue models， validated in a mouse model Leung et al.［82］ Inhomogeneous medium Simulate ultrasound-modulated optical tomography， 70 times faster Cai et al.［83］ Not mentioned Fast perturbation method on GPU［53］， 1000 times faster Young-Schultz et al.［85］ Inhomogeneous medium Develope FullMonteCUDA based on FullMonte［84］， 288‒936 times faster Franciosini et al.［86］ 3D voxel model Develope FRED for dose evaluation in radiation therapy， 200 to 5000 times faster Fang et al.［87］ Tetrahedral mesh model，inhomogeneous tissue Supporte complex light sources， wide-field detectors， and photo relay； 420 times faster Networked computers Kirkby et al.［78］ Semi-infinite medium Accelerate MC simulation on multiple networked computers Pratx et al.［79］ Inhomogeneous medium Large-scale parallel cloud computing， 1258 times faster on a 240 nodes cluster Doronin et al.［88］ Inhomogeneous medium Develop a peer-to-peer （P2P） program， 3 times faster than GPU simulation
Table 3. Comparison of different acceleration methods
View tableTable 3. Comparison of different acceleration methods
Method Acceleration factor Relative error Advantage Disadvantage Baseline fitting method Approximately 200 times［ 91 ］Less than 4%［ 91 ］Relatively low error Require storage of baseline simulation results， suitable only for layered tissues Perturbation method Approximately 1000 times［ 53 ］Less than 4%［ 53 ］Suitable for complex tissue photon transport simulation Limited to small differences between simulated and baseline tissues Hybrid method Approximately 300 times［ 62 ］Less than 5%［ 62 ］Flexible method selection based on optical parameters of tissues Only suitable for specific homogeneous tissue models Variance reduction method Approximately 300 times［ 58 ］Less than 5%［ 58 ］Flexible method selection Speed is limited by specific algorithms CPU multithreading method Approximately 1000 times［ 75 ］No error［ 75 ］Low error， strong universality High power consumption GPU multithreading method Approximately 5000 times［ 88 ］No error［ 88 ］Very high acceleration factor， high development potential Power consumption increases rapidly with speed Multi-device method Approximately 10000 times［ 88 ］No error［ 88 ］Low error， acceleration scales linearly with the number of devices， can be combined with other methods Acceleration limited by device interconnect and storage speed Hardware acceleration method Approximately 80 times［ 90 ］No error［ 90 ］High energy efficiency Limited application field， fewer research teams
Tools
Get Citation
Copy Citation Text
Louzhe Xu, Ting Li. Comprehensive Review of Acceleration Techniques for Monte Carlo Simulations of Photon Transport in Biological Tissues[J]. Chinese Journal of Lasers, 2024, 51(21): 2107104
Paper Information
Category: Biomedical Optical Imaging
Received: Feb. 26, 2024
Accepted: May. 15, 2024
Published Online: Oct. 31, 2024
The Author Email: Li Ting (t.li619@foxmail.com)
CSTR:32183.14.CJL240615
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20