Comprehensive Sensitivity Analysis for Particle Therapy Assessing Uncertainties in Range, Motion and Biological Modeling

质子和碳离子是最常用和讨论的治疗相关颗粒。由于其物理特性（布拉格峰，横向扩展）和由此产生的陡峭的剂量梯度，治疗结果是更敏感的运动和范围的不确定性比传统的X射线治疗的情况。进一步的不确定性源于生物建模，这是必要的，因为这些颗粒在细胞杀伤中更有效。它们的相对生物有效性（RBE）必须通过基于X射线参考数据的生物模型来预测或外推。在临床实践中，只有治疗计划者的知识和专业知识才能解释这些不确定性。新的发展需要为全面的风险评估提供客观的工具，到目前为止，还没有尝试将联合收割机范围、运动和生物建模的不确定性结合到粒子治疗的全面灵敏度分析中。该项目旨在建立这种全面的敏感性分析，引入统计（蒙特卡罗）敏感性分析方法，基于每个患者病例的大量（10000-100000）治疗计划计算。随机数分布解释了不同的不确定性。统计评估提供了新的信息来源，因为它会产生所有常见治疗计划评估可能性的平均值和相应方差（误差条）。这包括例如基于体素的三维方差（误差条）图、集成到剂量-体积直方图中的误差条以及肿瘤控制概率（TCP）、正常组织并发症概率（NTCP）和等效均匀剂量（EUD）值的置信区间。敏感性分析提供了不同不确定性的重要性的排名。在这个项目中，大量的治疗计划计算在图形处理单元（GPU）上并行化。现有的粒子治疗计划工具将有效地实施，以提供快速并行化，定制的灵敏度分析的要求。开发的源代码将上传到一个开放源代码库。该项目提供了新的见解粒子治疗的不确定性的组成，以及到他们的起源，并可用于治疗计划过程中的综合风险评估。目标是调查每个不确定性（范围，运动或生物建模）对治疗计划的重要性。开发的软件的未来应用包括患者特定的利润概念，基于灵敏度信息或鲁棒的优化方法，使用的期望值（平均值）和它们的方差。

Protons and carbon ions are the most used and discussed therapeutically relevant particles. Due to their physical properties (Bragg peaks, lateral spread) and the resulting steep dose gradients, treatment outcome is more sensitive to motion and range uncertainties than it is the case with conventional x-ray treatment. Further uncertainty originates from the biological modeling which is necessary because these particles are more efficient in cell killing. Their relative biological effectiveness (RBE) has to be predicted or extrapolated by biological models based on x-ray reference data. In clinical practice, solely the knowledge and expertise of the treatment planer account for these uncertainties. New developments need to provide objective tools for a comprehensive risk assessment.Up to now, there has been no attempt to combine range, motion and biological modeling uncertainties into a comprehensive sensitivity analysis for particle therapy. This project aims to establish this comprehensive sensitivity analysis, introducing a statistical (Monte Carlo) sensitivity analysis approach, based on a large number (10000-100000) of treatment plan calculations for every patient case. Random number distributions account for different uncertainties. A statistical evaluation provides new sources of information, as it results in mean values and corresponding variances (error bars) for all common treatment plan evaluation possibilities. This includes, for example, voxel-based, three-dimensional variance (error bar) maps, error bars integrated into dose-volume histograms and confidence intervals for tumor control probabilities (TCP), normal tissue complication probabilities (NTCP) and equivalent uniform dose (EUD) values. The sensitivity analysis provides a ranking of the significance of different uncertainties.In this project, the large number of treatment plan calculations is parallelized on graphical processing units (GPUs). Existing treatment planning tools for particle therapy will be efficiently implemented to provide fast parallelization, tailored to the requirements of the sensitivity analysis. The developed source code will be uploaded to an open source repository. This project provides new insight into the composition of uncertainties in particle therapy as well as into their origin and can be used for a comprehensive risk assessment in the treatment planning process. The goal is to investigate, how crucial each uncertainty (in range, motion or biological modeling) is for the treatment plan. Future applications of the developed software include patient specific margin concepts, based on the sensitivity information or a robust optimization approach that uses the expectation (mean) values and their variances.