RapidPlan精炼模型方法在宫颈癌中的应用

A new approach to model refinement in RapidPlan for predicting intensity-modulated radiotherapy plans for cervical cancer

目的:提出一种RapidPlan精炼模型的新方法,试图减少建模的工作量以及提高模型的质量和预测能力,并验证该精炼模型方法在宫颈癌病例中的应用。方法:随机选取20个已经批准并完成IMRT计划治疗的宫颈癌病例,导入RapidPlan数据库并建立DVH预测模型,定义为初始模型。然后利用初始模型重新预测20个病人的治疗计划,得到20个新计划,并导入新的数据库以此建立新的DVH预测模型,称之为精炼模型,并将该模型确定为特定病种的最终模型。为了明确在宫颈癌模型中精炼模型是否比初始模型有更好的预测能力以及能预测出更优的结果,通过比较两个模型的训练参数并分别用两个模型预测新入选的10个训练库以外的已经批准并治疗完成的IMRT计划,比较两组计划的DVH图和相关结构的计划质量指标V_(30)、V_(40)、D_(mean)。结果:在宫颈癌模型中精炼模型训练参数优于初始模型,两种模型预测计划的靶区适形度和均匀性差异均无统计学意义(P>0.05),且精炼模型预测的新计划中危及器官剂量低于初始模型,且差异具有统计学意义(P<0.05)。结论:该精炼模型的方法应用在宫颈癌中能提高模型质量和模型预测能力,该方法在减少宫颈癌RapidPlan制作的工作量的同时,大大改善了RapidPlan模型预测的结果。

Objective To propose a new method for refining the model in RapidPlan to reduce the workloads and improve the model＇s quality and performance in predicting the treatment plans of intensity- modulated radiotherapy（IMRT） for cervical cancer. Methods The data of 20 randomly selected cases of cervical cancer with approved and completed IMRT planning were imported into RapidPlan database to construct a dose- volume histogram（DVH） prediction model, which was defined as the initial model. Using the initial model, the IMRT plans of the patients were re- predicted to obtain 20 new plans, which were input into the new database to develop a new DVH prediction model to serve as the refined model and also as the final model for a particular disease. The quality and predictive ability of the refined model were compared with those of the initial model.The major metrics in the training logs of the two models were compared, and the DVH and the plan quality parameters V30,V40,and Dmeanpredicted by the two models were analyzed using additional 10 IMRT plans for cervical cancer. Results The refined model showed better training parameters than the initial model. No significant difference was found in the heterogeneity index and conformal index of the target area in the predicted treatment plans between the two models（P〈0.05）. The dose of organsat-risk（OAR） was significantly lower in the refined model than in the initial model（P〈0.05）. Conclusion The refined model shows improved model quality and predictive ability of IMRT plans for cervical cancer, and greatly improves the prediction results of RapidPlan model with decreased workloads.