基于预测剂量引导的宫颈癌自动计划研究

目的:使用机器学习方法建立宫颈癌计划剂量预测回归模型，并将预测剂量引导生成Monaco 计划系统（TPS）可调用的优化模板文件，实现宫颈癌的自动计划设计。方法:对50例宫颈癌术后调强治疗计划中的危及器官采集基于重叠体积直方图的几何特征值和基于剂量直方图的剂量目标值，建模后将模型预测剂量结果自动生成Monaco TPS模板文件，进而由TPS调用优化。使用该方法对另外10例未参与模型训练的测试病例进行自动计划设计，并和人工设计的计划进行对比分析。结果:自动计划比手动计划的平均设计时间减少了40 min（P<0.05），且平均调优次数降低了3次（P<0.05），剂量学指标和计划执行效率上两者无明显差异（P>0.05）。结论:基于预测剂量引导的宫颈癌自动计划可以达到临床要求，并且提高了计划设计效率。

Automatic planning of radiotherapy for cervical carcinoma based on dose prediction

Objective To establish a regressionmodel for cervical cancer planning dose prediction usingmachine learningmethod,and to realize the automatic planning of radiotherapy for cervical cancer by guiding the predicted dose to generate an optimizationtemplate file that can be called byMonaco treatment planning system (TPS).Methods The geometric characteristic values basedon overlap volume histogram and the dose target values based on dose volume histogram of organs-at-risk in the postoperativeintensity-modulated radiotherapy plans of 50 cervical cancer patients were collected.Aftermodeling, the dose predicted by themodelwas automatically generated into aMonaco TPS template file which was then optimized and called by TPS. The proposedmethodwas used for automatic planning in 10 test cases that did not participate inmodel training, and the obtained plans were then comparedwith themanually designed plans. Results Compared with those ofmanual planning, the average design time of automatic planningwas reduced by 40min (P<0.05), and there were 3 optimization times less in automatic planning (P<0.05). No significant differencewas found in dosimetry indexes and plan execution efficiency (P>0.05). Conclusion The automatic planning of radiotherapy forcervical cancer based on dose prediction canmeet clinical requirements and improve planning efficiency.