Abjective: Real-time 3D tumor location via monoscopic X-ray imaging is an attractive approach because it can provide accurate treatment and reduce the additional dose exposing to the patients without adding extra equipments. The key problem of the location algorithms is the insufficient information along the direction of imaging x-ray. Methods: Four representative location algorithms are selected for comparison using the simulated respiratory motion in this study. There are α-map algorithm, two algorithms based on 3D Gaussian probability density and Bayesian algorithm. Results: Treatment consisting of 15 fractions is simulated in this study. The 99th percentile and max error of the a-map algorithm are most greater than lcm. The 95th per-centile, 99th percentile and max error of the Gaussian algorithm No.1 are all greater than 1 cm. There are outliers ranged from 3 cm to 4 cm appearing in the max error of the Gaussian algorithm No 2. And the Bayesian algorithm has better results. If the model parameter p is less than 1, the ranges of rmse (root mean square error), 95th percentile, 99th percentile and max error are 1.8 mm～2 mm,3.4 mm～4.3 mm、4.2 mm～6 mm and 4.8 mm～6.5 mm respectively. Conclusions: Compared with the first three algorithms, the bayesian algorithm can provide more accurate and stable real-time tumor motion location for simulated respiratory motion.%目的:通过单射野方向的X射线成像来实时定位肿瘤靶区三维位置,不仅能够提高治疗精度,减少成像装置,而且能降低对病人曝光的额外剂量.但是由于射线束方向的运动信息缺失,如何保证高精度是定位算法需解决的一个关键问题.方法:本研究针对4种典型的实时定位算法,使用模拟的呼吸运动轨迹来比较它们的定位效果.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.结果:在连续15分次的模拟结果中,α分布图法的99百分位数(99th percentile)与最大误差绝大多数大于1 cm,高斯概率密度法1的95th percentile、99th percentile与最大误差均大于1 cm,高斯概率密度分布法2的最大误差项中出现了3 cm～4 cm的异常值,只有贝叶斯概率密度分布法的结果均较为稳定,在模型参数P＜1的情况下,均方根误差、95th percentile、99th percentile和最大误差分别为1.8 mm～2 mm、3.4 mm～4.3 mm、4.2 mm～6 mm和4.8 mm～6.5 mm.结论:相较于前3种算法,贝叶斯概率密度分布法总体上能够对模拟的呼吸运动引起的靶区运动实现更加精确和稳定的实时定位.
展开▼
