Acceleration techniques for SimSET SPECT simulations

摘要 发射层析成像仿真系统（SimSET）是发射层析成像的基础工具之一 断层扫描研究，被全世界数百名研究人员用于正电子发射断层扫描， (PET)和单光子发射计算机断层扫描（SPECT）。它被证明是准确和有效的 适用于PET和低能SPECT研究;因为SimSET使用几何模型进行SPECT 准直，它是不太准确的高能量同位素。本申请提出通过使用 角响应函数（ARF），这是一种已被证明可以准确建模SPECT准直的技术， 探测高能同位素比全光子跟踪模拟更有效。另外我们 提出了一种新的基于自回归函数的重要性采样方法，将这些模拟的速度提高了50倍以上。 ARF表的生成是另一个需要考虑的问题：它是计算密集型的， 仅在需要使用相同同位素/准直器/探测器进行大量模拟时使用 组合.为此，我们还提出了应用重要性采样来加速生成 ARF表的一个因素5，并建立了一个库的角度响应函数的流行 同位素/准直器/探测器组合。前者将减少生成 对于许多用户/用途来说，后者将完全消除生成ARF表的需要。这将大大 扩展了基于自回归函数的模拟的潜在应用。 我们的第一个目标是在不牺牲准确性的情况下加速SimSET SPECT模拟。这将是 通过协同使用两种工具完成：方差缩减和角度响应函数（ARF） tables.方差缩减包括重要性抽样和强制检测。我们假设这些 技术结合信息从我们的角度响应函数表将改善SimSET 模拟效率是特定放射性同位素的SPECT模拟的>50倍（例如，I-123、Y-90等）。我们 第二个目标是加速ARF表的生成。这将通过使用重要性抽样来实现 ARF的生成方法。我们还建议使用自适应分层方案， 仅在需要确定其值到用户指定值时才模拟给定表位置的光子 精度我们的第三个目标是为流行的供应商创建一个预先计算的ARF表库 同位素/准直器/探测器配置。这些ARF表格将被提供给 通过SimSET网站下载。凭借超过500名注册用户，我们相信， SimSET的增强将对世界各地的研究项目产生深远的影响。

Abstract The Simulation System for Emission Tomography (SimSET) is one of the foundational tools for emission tomography research, used by hundreds of researchers worldwide for both positron emission tomography (PET) and single photon emission computed tomography (SPECT). It has proven to be accurate and efficient for both PET and low energy SPECT studies; because SimSET uses a geometric model for its SPECT collimation, it is less accurate for high energy isotopes. This application proposes to address this with the use of angular response functions (ARFs), a technique that has proven to accurately model SPECT collimation and detection for high-energy isotopes more efficiently than full photon-tracking simulations. In addition, we propose a novel ARF-based importance sampling method that will speed these simulations by a factor of >50. The generation of ARF tables is another consideration: it is extremely compute intensive and has caused ARF to be used only when a large number of simulations are needed using the same isotope/collimator/detector combination. For this reason, we also propose application of importance sampling to speed the generation of ARF tables by a factor 5, and the creation of a library of angular response functions for popular isotope/collimator/detector combinations. The former will lessen the computational cost of generating the tables, the latter will, for many users/uses, eliminate the need to generate ARF tables at all. This will greatly expand the potential applications of ARF-based simulations. Our first aim is to accelerate SimSET SPECT simulations without sacrificing accuracy. This will be accomplished by synergistically utilizing two tools: variance reduction and angular response function (ARF) tables. Variance reduction includes importance sampling and forced detection. We hypothesis that these techniques combined with information from our angular response function tables will improve SimSET simulation efficiency by >50 times of SPECT simulations of specific radioisotopes (e.g., I-123, Y-90, etc.). Our second aim is to accelerate ARF table generation. This will be accomplished by using importance sampling methods in the generation of ARFs. We further propose to use an adaptive stratification scheme that will simulate photons for a given table position only as long as required to determine its value to a user-specified precision. Our third aim is to create a library of pre-calculated ARF tables for popular vendor isotope/collimator/detector configurations. These ARF tables will then be made publically available for download through the SimSET website. With a registered user base of >500, we believe that these enhancements to SimSET will have far reaching impact on research projects throughout the world.