瓦里安RapidPlan模型训练中统计离群值的处理及其剂量学影响

发布时间：2018-06-30 06:39

本文选题：Rapid + Plan　；参考：《中国医学物理学杂志》2016年07期

【摘要】：目的:以统计图表和参数为依据,分析处理瓦里安Rapid Plan模型离群值,并比较处理前后计划之间及其与人工优化的剂量学差异。方法:(1)选取80例直肠癌术前同步推量计划建立模型;(2)修正明显原因所致的离群值;(3)利用拟合曲线、Z值和库克距离寻找几何离群值,检查原计划并分情况处理;(4)利用残差分布图、剂量体积直方图、学生化残差值和库克距离判断剂量离群值并酌情处理;(5)利用20例同类计划测试验证前后模型的优化效果。结果:统计学验证使模型构成计划中的离群参数最大值显著降低(P0.05)。相比原人工优化的测试计划,Rapid Plan使用任一模型均能显著改善靶区均匀性指数(P0.05,幅度2.7%~19.0%)和膀胱平均剂量(P0.05,降幅12.8%~13.2%)。但利用验证前后模型分别优化的计划质量差异不大:肿瘤计划靶区(PGTV)均匀性指数相差0.5%,PGTV适形指数相差0.1%,计划靶区适形指数相差0.5%,股骨头和膀胱平均剂量分别相差0.3%和0.4%(P0.05)。结论:基于知识的Rapid Plan计划可以改善靶区剂量均匀性和保护危及器官。依据统计学参数删改模型构成计划并不一定能取得更好的剂量学效果。
[Abstract]:Aim: based on statistical charts and parameters, the outliers of Varian Rapid Plan model were analyzed, and the dosimetric differences between plans and artificial optimization were compared before and after treatment. Methods: (1) 80 cases of rectal cancer were selected to set up a model with synchronous push plan before operation; (2) outliers caused by obvious causes were corrected; (3) geometric outliers were found by using fitting curve Z value and Cook distance, and the original plan was checked and processed by case; (4) residual distribution map was used. Dose volume histogram, student-based residuals and Cook distance were used to judge the dose outliers and to deal with them as appropriate. (5) the optimization effect of the model was verified by 20 similar programs. Results: statistical verification significantly reduced the maximum value of outliers in the model planning (P0.05). Compared with the original artificial test plan, Rapid Plan could significantly improve the target area uniformity index (P0.05, range 2.7U 19.0%) and the average bladder dose (P0.05, decrease 12.8m 13.2%) using any model. The difference of PGTV homogeneity index (PGTV) was 0. 5%, PGTV conformability index was 0. 5% and 0. 5%, and the average dose difference of femoral head and bladder was 0. 3% and 0. 4% respectively (P0.05). Conclusion: Knowledge-based Rapid Plan can improve the dose uniformity of target area and protect endangered organs. The formulation of the model based on statistical parameter deletion does not necessarily achieve better dosimetry results.
【作者单位】：北京大学肿瘤医院暨北京市肿瘤防治研究所放疗科恶性肿瘤发病机制及转化研究教育部重点实验室;
【基金】：国家自然科学基金(11505012,81402535) 北京市医院管理局“青苗”计划专项(QML20151004) 质检公益性行业科研专项(201510001-02)
【分类号】：R730.55

【相似文献】