【作者】 周正东；

【导师】 舒华忠；

【作者基本信息】 东南大学 ， 生物医学工程， 2005， 博士

【摘要】 优化是放射治疗计划系统的重要组成部分之一。针对放射治疗方案的优化问题,本文提出了一系列解决方案,主要包括以下几个方面:1)较全面地介绍了现代最优化理论及方法,对NSGA-II多目标优化遗传算法进行了较深入的研究,提出了在该方法中加入局部搜索算子及外部精英收集池的改进算法-NSGA-II多目标混合遗传算法。2)针对伽玛刀立体定向放射治疗计划系统中的优化问题提出了一个两步优化的方案。首先采用一种改进的基于距离变换的几何优化方法,优化靶点的数目、位置及相应准直器直径;然后根据几何优化获得的靶点参数,研究了基于混合遗传算法GA-BFGS方法优化靶点权重的方法,以避免梯度法可能会陷入局部极小的问题。3)针对全身伽玛刀治疗体部肿瘤的靶点设计优化问题,提出了一种新颖的靶点设计及优化解决方案,建立了基于包围盒的优化模型,然后采用遗传算法进行求解,使治疗过程更为简便快捷,并有利于减少治疗过程中的定位误差,填补了全身伽玛刀靶点设计优化研究的空白。4)针对全身伽玛刀治疗体部肿瘤的治疗路径优化问题,在靶点优化的基础上,提出了采用遗传算法进行治疗路径优化的解决方案,使治疗过程更为简便快捷,有利于减少治疗过程中医生的劳动强度。当靶点数过多时,我们给出了一种分层优化的解决方案,以减少优化时间,满足临床的实际要求。5)实现了一种基于NSGA-Ⅱ多目标混合遗传算法的调强放射治疗逆向计划的优化方法,其中局部搜索采用L-BFGS算法。该方法得到的非劣解在目标空间分布均匀,计算速度快,鲁棒性好。6)在Windows平台上开发了一套用于周围血管内放射治疗计划优化的主动漫游虚拟现实系统,用于提取血管的几何结构;建立血管内放射治疗计划的优化模型,采用适宜的优化方法进行优化,该方法综合运用了混合遗传算法GA-BFGS方法和快速模拟退火算法优化相关的参数,其中GA-BFGS算法用于优化连续参数即放射源的照射时间,快速模拟退火算法用于优化离散参数即放射源的位置。这一策略综合考虑了目标函数的形式,BFGS的快速收敛性,遗传算法的全局收敛性,以及模拟退火算法的全局收敛性。更多还原

【Abstract】 Optimization plays an important role in the radiation treatment planning. In this thesis, a collection of optimization solutions is presented for the radiation treatment planning problems.Firstly, modern optimization theory and methods are reviewed, and a hybrid multiobjective optimization algorithm, which combines state-of-the-art multiobjective optimization algorithm-NSGA-II with local search methods and external elitist pool, is developed. The experimental results show that hybridization with local search can efficiently improve the search ability of the multiobjective optimization algorithms, and well distributed Pareto set could be got.Secondly, a two-step automated treatment planning algorithm is developed for the Leksell Gamma Knife, a specialized unit for radiation treatment of brain tumour. In the first step, an improved distance transform based method is used to find the best number of shots, shot locations and collimator sizes for treatment planning. In the second step, in order to avoid trapping into local minimum, a hybrid genetic algorithm combined with BFGS algorithm is employed to find the optimal radiation exposure time by fixing the values of the discrete variables generated in the first step.Thirdly, a novel shot design and optimization method is proposed for the whole body Gamma Knife treatment planning system which is used to treat large tumor, a boundary box based optimization model is proposed and solved by genetic algorithm. The method makes the treatment procedure more convenient and fast, and reduces the shot location error during the treatment procedure.Fourthly, on the basis of shots optimization arrangement for the whole body Gamma Knife treatment planning system, genetic algorithm is applied to find the best treatment path to make the treatment procedure more convenient and reduce the doctor’s work intensity. In case of a large amount of shots are needed, a layer by layer treatment path optimization scheme is proposed to reduce computation time and meet the clinical need, but with a little performance decrease. Fifthly, NSGA-Ⅱhybrid multiobjective optimization algorithm,using L-BFGS algorithm as local search method, is applied to the optimization of inverse planning in intensity modulated radiation therapy(IMRT). The non-dominated solutions obtained by hybrid multiobjective optimization algorithm are distributed uniformly and this algorithm is more robust than those of the weighted method. The last non-dominated solutions allow the doctors to select the solution which best fits the clinical needs according to the corresponding decision tools such as dose-volume histograms, isodose lines, and the distribution of the solutions.Finally, a method combined the active navigation with optimization algorithm for the planning of peripheral intravascular brachytherapy is proposed. A virtual reality system based on active navigation is developed on Windows platform for the peripheral intravascular brachytherapy, and an optimization model is proposed for the peripheral intravascular brachytherapy. According to the characteristics of the vessel obtained by active navigation, a method combined GA-BFGS hybrid genetic algorithm with simulated annealing algorithm is applied to optimize the related parameters,GA-BFGS hybrid genetic algorithm is used to optimize the continuous parameters, i.e., the dwell time; simulated annealing is used to optimize discrete parameter, i.e., the dwell positions located at the centreline of the vascular. This strategy takes into consideration the formula of the objective function, the fast convergence of BFGS algorithm and the global convergence of both genetic algorithm and simulated annealing algorithm.更多还原