[论文解读] A conceptual study on real-time adaptive radiation therapy optimization through ultra-fast beamlet control

R Wiersma, Xinmin Liu|arXiv (Cornell University)|Nov 8, 2018

Advanced Radiotherapy Techniques参考文献 44被引用 5

一句话总结

本文提出了一种使用超快速射野子束控制的实时自适应放射治疗（RT-ART）框架，可动态响应治疗过程中患者的分次内运动，实现剂量递送的实时优化。通过模拟以100 Hz二值化MLC运行的快速旋转直线加速器，并在运动超过阈值时求解时间依赖的目标函数，RT-ART在各种运动类型下实现了几乎与理想静态情况相当的剂量适形度——PTV D95/V95从无自适应时的75.6%/71.8%提升至97.2%/96.7%。

ABSTRACT

A central problem in the field of radiation therapy (RT) is how to optimally deliver dose to a patient in a way that fully accounts for anatomical position changes over time. As current RT is a static process, where beam intensities are calculated before the start of treatment, anatomical deviations can result in poor dose conformity. To overcome these limitations, we present a simulation study on a fully dynamic real-time adaptive radiation therapy (RT-ART) optimization approach that uses ultra-fast beamlet control to dynamically adapt to patient motion in real-time. A virtual RT-ART machine was simulated with a rapidly rotating linear accelerator (LINAC) source (60 RPM) and a binary 1D multi-leaf collimator (MLC) operating at 100 Hz. If the real-time tracked target motion exceeded a predefined threshold, a time dependent objective function was solved using fast optimization methods to calculate new beamlet intensities that were then delivered to the patient. To evaluate the approach, system response was analyzed for patient derived continuous drift, step-like, and periodic intra-fractional motion. For each motion type investigated, the RT-ART method was compared against the ideal case with no patient motion (static case) as well as to the case without the use RT-ART. In all cases, isodose lines and dose-volume-histograms (DVH) showed that RT-ART plan quality was approximately the same as the static case, and considerably better than the no RT-ART case. The RT-ART optimization framework has the potential to optimally deliver dose to a patient in a way that fully accounts for anatomical changes due to motion. With continued advances in real-time patient motion tracking and fast computational processes, there is significant potential for the RT-ART optimization process to be realized on next generation RT machines.

研究动机与目标

• 解决静态放射治疗计划在治疗过程中无法考虑分次内解剖结构变化的局限性。
• 开发一种实时自适应优化框架，根据追踪到的患者运动动态调整射野子束强度。
• 评估实时射野子束控制是否能在存在运动的情况下，维持与理想静态计划相当的高剂量适形度。
• 展示利用先进直线加速器配置实现超快速优化与剂量交付的可行性。

提出的方法

• 模拟了一个虚拟的RT-ART系统，采用60 RPM旋转的直线加速器和100 Hz二值化一维多叶准直器（MLC）。
• 实现了一个时间依赖的目标函数，当实时追踪到的靶区运动超过预设阈值时触发优化。
• 采用快速优化方法实时计算新的射野子束强度，实现对运动的动态适应。
• 采用标准TG-119体模，并测试了三种运动类型：阶跃式、连续漂移和周期性运动。
• 假设实时可变形图像配准（DIR）将提供靶区和危及器官（OAR）运动的输入，但本研究中未实际实现该功能。
• 由于所提出的配置中未考虑机架和MLC速度约束，采用凸优化，确保全局最优解。

实验结果

研究问题

• RQ1 在各种分次内运动模式下，实时射野子束控制能否维持与理想静态计划相当的剂量适形度？
• RQ2 在运动引起的解剖结构位移下，RT-ART相对于传统静态计划和理想无运动情况的表现如何？
• RQ3 超快速优化与交付对在动态照射中实现临床可接受的剂量适形度有何影响？
• RQ4 所提出的RT-ART框架在当前或近期内的直线加速器硬件与计算能力下是否具备技术可行性？

主要发现

• RT-ART将无自适应情况下的PTV D95/V95从75.6%/71.8%提升至所有运动类型下的97.2%/96.7%。
• RT-ART实现的剂量适形度几乎与理想静态情况完全一致，后者PTV D95/V95为96.8%/97.0%。
• 该方法有效恢复了阶跃式、连续漂移和周期性运动下的剂量适形度，表明其在各类运动模式下均具有鲁棒性。
• 所提出的系统配置支持凸优化，保证获得全局最小解，简化了实时计算。
• 超快速MLC控制（100 Hz）和快速机架旋转（60 RPM）足以实现实时射野子束访问和动态适应。
• 未来实时DIR和GPU加速优化的进展有望进一步缩短处理时间，提升临床可行性。