基于快速梯度下降的容积旋转调强优化算法

Fast gradient descent based VMAT optimization algorithm

目的 针对目前容积旋转调强放疗（VMAT）计划中的优化时间长和解的可重复性差等问题，探讨一种基于快速梯度下降的优化算法。方法 利用梯度下降算法求解由少数固定野组成的常规调强放疗（IMRT）计划，通过叶片序列生成算法得到最优射野孔径形状和权重。在保持已有优化射野的前提下，渐进地增加并优化新的射野直到达到所需射野数。通过实际病例对该方法的性能进行评估。结果 针对头颈部肿瘤病例，VMAT计划的优化时间约为5 min，而目前使用的商业VMAT优化算法一般需要10～20 min。由于梯度下降算法为确定性算法，得到的优化解可重复。VMAT计划的靶区适形度和均匀度皆优于IMRT计划。对大部分危及器官的保护而言，VMAT计划略好于IMRT计划。结论 和已有VMAT优化算法相比，新算法不仅优化时间大幅缩短，而且保证了解的可重复性。

Objective To explore a new gradient descent based optimization algorithm in order to improve the efficiency and unrepeatability of current VMAT planning systems. Methods Firstly, fast monotonic descent method was used to generate intensity maps of a simple IMRT plan consisting of few static fields. Then, leaf sequencing algorithm was employed to determine the best segments （aperture shape and weight） of these fields. The segments of the existing beams were fixed, and the new beams were continuously added to the plan and optimized until the maximal number of beams arrived. The performance of this algorithm was evaluated with clinical cases. Results For the head-and-neck case, the computation time was about 5 min while the time for those commercial planning systems was 10-20 min. Due to the nature of gradient-based algorithm, the repeatability of optimization result was guaranteed. The conformity and homogeneity of VMAT plan was better than that of IMRT plan. For most of critical organs, the dose sparing of VMAT plan was better than that of IMRT plan. Conclusions Compared to existing VMAT optimization algorithms, the computation time of our algorithm is significantly reduced and the optimization result is repeatable.