单方向X射线成像的肿瘤定位算法比较

发布时间：2018-03-29 23:04

本文选题：靶区运动　切入点：实时定位　出处：《北京协和医学院》2013年硕士论文

【摘要】：由于人体解剖结构的动态特性,病人胸腹部的器官和肿瘤靶区在治疗过程中常会发生运动,计划设计时适形的剂量分布在治疗时可能会偏离靶区,从而影响靶区剂量分布的适形性。为了降低放射治疗过程中靶区运动所带来的剂量不确定性,需要实施一定的靶区运动管理方法。目前提出的多种运动管理方法包括PTV包含运动范围、呼吸屏气治疗、门控治疗和实时跟踪治疗等等。但是前三种运动管理方法都有其一定的局限性；而实时跟踪治疗作为一种最精确的运动管理方法,如何精确地实时定位肿瘤靶区运动是这种方法最大的挑战。目前已提出了多种肿瘤靶区实时定位方法,主要可分为直接定位法、间接定位法和混合定位法三大类。直接定位法中,根据是否使用X射线成像的情况又分为X射线成像定位(包括立体X射线成像和单方向X射线成像等)和非X射线成像定位(包括电磁定位、超声定位和核磁加速器在机MRI定位等)；间接定位法主要包括了利用体外标记物、体外压力传感器、体表轮廓变化等进行体表运动监测和利用肺活量计等测量呼吸容量来进行辅助定位的方法；混合定位法则将前两种方法结合应用,典型产品是射波刀(CyberKnife)和ExacTrac同步跟踪定位系统。相比较而言,单方向X射线成像定位法应用简单,而且能够有效降低对病人曝光的成像剂量,如果使用合适的定位算法,则能够方便有效地应用于临床。因此,本研究选取了4种典型的单方向X射线成像定位算法,使用模拟的呼吸运动轨迹和前列腺肿瘤病人的Calypso跟踪数据来比较它们的定位结果。这四种定位算法分别是α分布图法、两种基于高斯概率密度分布的算法和贝叶斯概率密度分布法。在连续15分次对模拟的呼吸运动轨迹的模拟结果中,α分布图法的均方根误差范围约为2.8-4.6mm,而最大误差的范围为12.9-42.3mm；高斯概率密度法1的均方根误差范围约为6.5-8.2mm,最大误差范围为12.8-15.3mm；高斯概率密度分布法2的均方根误差范围约为1.6-3mm,在绝大多数分次中最大误差范围为4.4-6.1mm,但也出现了超过30mm的异常值；贝叶斯概率密度分布法的均方根误差范围约为1.8-2mm,而最大误差范围为4.8-6.5mm。在对10位前列腺肿瘤病人的Calypso跟踪数据的模拟结果中,α分布图法的均方根误差范围约为0.2-5.1mm,而最大误差的范围为0.7-55.7mm；高斯概率密度法1的均方根误差范围约为0.2-2.6mm,最大误差范围为0.7-5.9mm；对于绝大多数病人,高斯概率密度分布法2的均方根误差范围约为0.15-1.4mm,最大误差范围为0.5-7mm,但有少数病人的均方根误差为4.5-8.4mm,最大误差超过了30mm；贝叶斯概率密度分布法的均方根误差范围约为0.15-2.5mm,而最大误差范围为0.5-8.8mm。相比较而言,贝叶斯概率密度分布法能够最好地适用于呼吸运动引起的肿瘤运动和多种类型的前列腺肿瘤运动的实时定位。
[Abstract]:Because of the dynamic characteristics of the anatomical structure of the human body, the organs and tumor targets of the patient's chest and abdomen often move during the course of treatment, and the conformal dose distribution of the planned design may deviate from the target area during the treatment. Thus the conformability of dose distribution in target area is affected. In order to reduce the dose uncertainty caused by target motion during radiotherapy, a certain method of target motion management should be implemented. At present, a variety of motion management methods, including PTV, including range of motion, breath-holding therapy, are proposed. But the first three methods of motion management have some limitations, while real-time tracking therapy is the most accurate method of motion management. How to accurately locate tumor target motion in real-time is the biggest challenge of this method. At present, a variety of real-time localization methods for tumor target have been proposed, which can be divided into three categories: direct localization method, indirect localization method and mixed localization method. Depending on whether X-ray imaging is used, it is subdivided into X-ray imaging positioning (including stereoscopic X-ray imaging and unidirectional X-ray imaging, etc.) and non-X-ray imaging positioning (including electromagnetic positioning). Ultrasonic localization and nuclear magnetic accelerator positioning in the machine MRI; indirect localization mainly includes the use of external markers, in vitro pressure sensor, The methods of surface motion monitoring and breathing volume measurement such as vital capacity meter are used to locate the body surface contour, and the hybrid localization method combines the first two methods. The typical products are CyberKnifeand ExacTrac synchronous tracking and positioning system. In comparison, the single direction X-ray imaging localization method is simple to use and can effectively reduce the imaging dose of patient exposure, if the appropriate localization algorithm is used, Therefore, four typical single-direction X-ray imaging localization algorithms are selected in this study. The simulated respiratory track and Calypso tracking data of prostate cancer patients were used to compare their localization results. The four localization algorithms are alpha distribution method. Two algorithms based on Gao Si probability density distribution and Bayesian probability density distribution method. The mean square error range of 伪 distribution map method is about 2.8-4.6mm, and the maximum error range is 12.9-42.3mm, and that of Gao Si's probability density method 1 is about 6.5-8.2mm, the maximum error range is about 6.5-8.2mm. The error range is 12.8-15.3mm, the root mean square error range of Gao Si probability density distribution method 2 is about 1.6-3mm, and the maximum error range is 4.4-6.1mm in most grades, but there are outliers over 30mm. The root mean square error range of Bayesian probability density distribution method is about 1.8-2mm and the maximum error range is 4.8-6.5mm. In the simulation of Calypso tracking data of 10 prostate cancer patients, the root mean square error range of 伪 distribution map method is about 0.2-5.1mm, while the maximum error range is 0.7-55.7mm, and the RMS error range of Gao Si's probability density method 1 is about 0.2-2.6mm, the maximum error range is 0.7-55.7mm, and the maximum RMS error range is about 0.2-2.6mm for Gao Si's probability density method 1. The wide error ranges from 0.7 to 5.9 mm; for the vast majority of patients, The root-mean-square error range of Gao Si probability density distribution method 2 is about 0.15-1.4mm, the maximum error range is 0.5-7mm, but the RMS error of a few patients is 4.5-8.4mm, the maximum error is more than 30mm, and the RMS error range of Bayesian probability density distribution method is about 30mm. The maximum error range is 0.5-8.8mm. In contrast, Bayesian probabilistic density distribution method is the best method for real-time localization of tumor movements caused by respiratory movement and various types of prostate tumor movements.
【学位授予单位】：北京协和医学院
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：R730.44;TP391.41

【参考文献】