目的:研究在复杂处方和解剖分段等临床条件下，基于瓦里安RapidPlan模块建立食管癌放疗自动计划整合模型的可行性及其剂量学特点。方法:收集301例多处方、多分段的食管癌放疗计划导入RapidPlan系统，在ModelAnalytics(MA)工具辅助下对模型进行统计学分析和离群值处置，建立初始模型。再利用未参与建模的40例临床食管癌治疗计划，通过RapidPlan模型对其进行重新优化以评估模型的自动优化效果，并以此作为反馈数据，进一步调试和优化迭代模型参数，分别对两组计划的各主要参数进行比较。结果:通过增大样本量和按剂量比例(而非命名规则)匹配结构等方法，本工作成功建立了复杂处方和解剖分段下的整合RapidPlan食管癌自动放疗计划模型。RapidPlan和临床计划对于不同器官的保护各有优势，但均符合临床要求。相对传统人工计划的试错过程，RapidPlan更加高效，且减少了主观因素差异造成的计划质量不一致。结论:本工作提出了在复杂解剖分段和处方条件下建立食管癌自动放疗计划整合RapidPlan模型的方法，并基于独立病例验证了其可行性和剂量学表现。“,”Objective:To study the feasibility and dosimetric characteristics of establishing a comprehensive model for automated treatment planning for esophageal cancer based on Varian RapidPlan module under complex conditions such as different prescriptions and anatomical sections.Methods:In total, 301 historical plans with multi-prescription and multi-sectional esophageal cancer were imported into RapidPlan system. Assisted by the ModelAnalytics(MA) tool, statistical verification was performed to profess outliers, yielding the initial model; Additional 40 clinical esophageal cancer treatment plans were duplicated as validation set. The RapidPlan-based re-optimization result was assessed and used as feedback data to fine-tune the model parameters iteratively. The primary dosimetric parameters of the two groups were then compared.Results:Through enlarged training set sample size and structure matching (based on relative dose rather than nomenclature), a comprehensive model feasible of handling various anatomic sections and dose prescriptions was successfully established. Both clinical plans and RapidPlan re-optimization were clinically acceptable, displaying complementary dosimetric advantages. Compared with the trial-and-error process of conventional manual planning, RapidPlan method was more efficient and independent from subjective influence, which induced inconsistency of plan quality.Conclusions:This work proposed and validated a modeling method of automated treatment planning for esophageal cancer under complex anatomic section and dose prescription. Dosimetric performance of the model is assessed based on independent validation set.