综合复杂性特征和剂量学评估指标提高剂量验证结果预测模型的性能

目的:构建随机森林模型预测调强计划剂量验证结果,研究综合复杂性特征和剂量学评估指标提高模型性能的可行性。方法:选取269例IMRT计划,共2 558个射野,采用电子射野影像系统进行剂量验证,γ通过率(2%/2 mm标准)阈值为95%,将剂量验证结果分为“通过”和“不通过”。提取计划的剂量学评估指标和射野的复杂性特征,分别构建剂量模型(基于剂量学评估指标)、计划模型(基于计划复杂性特征)和混合模型(综合剂量学评估指标和计划复杂性特征)。计算AUC值、特异性和敏感性评估模型性能。结果:剂量模型、计划模型和混合模型的AUC值分别为0.68、0.80和0.82,混合模型优于其他两个模型。混合模型的特异性和敏感性为0.70和0.79,均高于其他两个模型。剂量模型、计划模型和混合模型达到最佳性能所需的样本量分别为1200、900和700。结论:剂量学评估指标与计划复杂性特征综合,可以提高模型的预测性能,同时在一定程度上弥补样本数量的不足,为预测剂量验证结果的机器学习模型性能的改善提供参考。

瓦里安RapidPlan模型训练中统计离群值的处理及其剂量学影响

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（P〈0.05）。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（P〈0.05,幅度2.7%~19.0%）和膀胱平均剂量（P〈0.05,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%（P〉0.05）。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。

Dosimetric study comparing photon and electron beams for boosting the tumor bed in early-stage breast cancer

Objective： The aim of our study was to assess and compare the potential dosimetric advantages and drawbacks of photon beams and electron beams as a boost for the tumor bed in superficial and deep seated early-stage breast cancer. Methods： We planned CTs of 10 women with early breast cancer underwent breast conservative surgery were selected. Tumor bed was defined as superficial and deep with a cut of point 4 cm, those with less than 4 cm were defined as superficial tumors representing 4 patients and those with depth of 4 cm or more were classified as deep tumors representing 6 patients. The clinical target volume （C＇I＇V） was defined as the area of architectural .distortion surrounded by surgical clips. The plan- ning target volume （PTV） was the C＇I＇V plus margin 1 cm. A dose of 10 Gy.in 2 Gy fractions was given concurrently at the last week of treatment. Organs at risk （OARs） were heart, lungs, contra-lateral breast and a 5 mm thick skin segment of the breast surface. Dose volume histograms were defined to quantify the quality of concurrent treatment plans assessing target coverage and sparing OARs. The following treatment techniques were assessed： photon beam with 3D-conformal technique and a single electron beam. Results： For superficial tumors better coverage for CTV and P＇I＇V with good homogeneity with better CI was found for the 3D conformal radiotherapy （3DCRT） but with no significant planning objectives over electron beam. For deep tumors, the 3DCRT met the planning objectives for C＇I＇V, PTV with better coverage and fewer hot spots with better homogeneity and CI. For superficial tumors, OARs were spared by both techniques with better sparing for the electron beam where as for deep tumors also OARs were well spared by both techniques. Conclusion： Boosting the tumor bed in early- stage breast cancer with optimized photon may be preferred to electron beam for both superficial and deep tumors. The OARs dose sparing effect may allow for a potential long-term toxicity risk reduction and better cosmesis.