Fang Qianqian
Martinos Center for Biomedical Imaging, Massachusetts General Hospital, 149 13th Street, Charlestown, Massachusetts, 02129, USA.
Biomed Opt Express. 2011 Apr 19;2(5):1258-64. doi: 10.1364/BOE.2.001258.
The Monte Carlo (MC) method is a popular approach to modeling photon propagation inside general turbid media, such as human tissue. Progress had been made in the past year with the independent proposals of two mesh-based Monte Carlo methods employing ray-tracing techniques. Both methods have shown improvements in accuracy and efficiency in modeling complex domains. A recent paper by Shen and Wang [Biomed. Opt. Express 2, 44 (2011)] reported preliminary results towards the cross-validation of the two mesh-based MC algorithms and software implementations, showing a 3-6 fold speed difference between the two software packages. In this comment, we share our views on unbiased software comparisons and discuss additional issues such as the use of pre-computed data, interpolation strategies, impact of compiler settings, use of Russian roulette, memory cost and potential pitfalls in measuring algorithm performance. Despite key differences between the two algorithms in handling of non-tetrahedral meshes, we found that they share similar structure and performance for tetrahedral meshes. A significant fraction of the observed speed differences in the mentioned article was the result of inconsistent use of compilers and libraries.
蒙特卡罗（MC）方法是一种用于模拟光子在诸如人体组织等一般浑浊介质中传播的常用方法。在过去一年里，基于光线追踪技术的两种基于网格的蒙特卡罗方法被独立提出，取得了进展。这两种方法在对复杂区域建模时，在准确性和效率方面都有改进。沈和王最近发表的一篇论文[《生物医学光学快报》2, 44 (2011)]报告了对这两种基于网格的MC算法和软件实现进行交叉验证的初步结果，显示两个软件包之间存在3至6倍的速度差异。在这篇评论中，我们分享我们对无偏软件比较的看法，并讨论其他问题，如预计算数据的使用、插值策略、编译器设置的影响、俄罗斯轮盘赌的使用、内存成本以及测量算法性能时的潜在陷阱。尽管这两种算法在处理非四面体网格方面存在关键差异，但我们发现它们在处理四面体网格时具有相似的结构和性能。上述文章中观察到的速度差异很大一部分是编译器和库使用不一致的结果。
