| 刘嘉城,王翰林,王清莹,等.剂量预测联合参数迭代优化算法的VMAT全自动计划研究[J].中华放射医学与防护杂志,2021,41(11):830-835.Liu Jiacheng,Wang Hanlin,Wang Qingying,et al.Fully automatic volumetric modulated arc therapy planning based on dose prediction combined with an iterative optimization algorithm[J].Chin J Radiol Med Prot,2021,41(11):830-835 |
| 剂量预测联合参数迭代优化算法的VMAT全自动计划研究 |
| Fully automatic volumetric modulated arc therapy planning based on dose prediction combined with an iterative optimization algorithm |
| 投稿时间：2021-06-08 |
| DOI：10.3760/cma.j.issn.0254-5098.2021.11.006 |
| 中文关键词: 自动计划 参数迭代优化算法 剂量预测 深度学习 直肠癌 |
| 英文关键词:Automatic planning Iterative optimization algorithm Dose prediction Deep Learning Rectal cancer |
| 基金项目:国家重点研发计划（2019YFF01014405）；国家自然科学基金（12005007）；北京市自然科学基金（1202009，1212011） |
|
| 摘要点击次数: 3858 |
| 全文下载次数: 1437 |
| 中文摘要: |
| 目的 设计一种联合深度学习剂量预测和参数迭代优化算法的容积调强放射治疗（VMAT）全自动计划方法。方法 选取2018年6月至2021年1月北京大学肿瘤医院既往165例直肠癌患者的VMAT计划开展研究，其中145例用于训练和验证深度学习模型，该模型用于预测危及器官的剂量，20例用于研究比对自动计划和人工计划的质量。该方法从危及器官的预测剂量分布中提取关键的剂量体积直方图（DVH）值作为初始优化参数（IOPs），利用治疗计划系统可编程接口自动创建VMAT计划，通过设计迭代优化算法自动调节优化参数（OPs）。结果 剂量预测模型训练后能有效预测出20例测试计划危及器官的关键DVH值，与参考值相比差异均无统计学意义（P>0.05）。20例VMAT自动计划均能满足临床处方剂量要求，对于PTV和PGTV的适形性指数（CI），人工计划与自动计划比较差异均无统计学意义（P>0.05）；而PGTV的D1和均匀性指数（HI），自动计划均高于人工计划，分别为0.6 Gy和0.01，两者比较差异均有统计学意义（t=-7.05、-6.92，P<0.05）。自动计划比人工计划的膀胱平均V30下降2.7%（t=3.37，P<0.05），股骨头和危及器官辅助结构（Avoidance）的平均V20分别下降8.37%和15.95%（t=5.65、11.24，P<0.05），并且膀胱、股骨头、Avoidance的平均剂量分别降低了1.91、4.01和3.88 Gy（t=9.29、2.80、10.23，P<0.05）。测试的20例直肠癌患者病例的自动计划平均时间为（71.82±25.48）min。结论 本研究利用直肠癌病例验证了一种联合剂量预测和参数迭代优化算法的VMAT自动计划方法的可行性。相比于人工计划，VMAT自动计划无需人工干预，在提高计划设计效率、计划质量和临床资源利用率等方面有很大的应用潜力。 |
| 英文摘要: |
| Objective To develope an automatic volumetric modulated arc therapy (VMAT) planning for rectal cancer based on a dose-prediction model for organs at risk(OARs) and an iterative optimization algorithm for objective parameter optimization. Methods Totally 165 VMAT plans of rectal cancer patients treated in Peking University Cancer Hospital & Institute from June 2018 to January 2021 were selected to establish automatic VMAT planning. Among them, 145 cases were used for training the deep-learning model and 20 for evaluating the feasibility of the model by comparing the automatic planning with manual plans. The deep learning model was used to predict the essential dose-volume histogram (DVH) index as initial objective parameters(IOPs) and the iterative optimization algorithm can automatically modify the objective parameters according to the result of protocol-based automatic iterative optimization(PBAIO). With the predicted IOPs, the automatic planning model based on the iterative optimization algorithm was achieved using a program mable interface. Results The IOPs of OARs of 20 cases were effectively predicted using the deep learning model, with no significantly statistical difference in the conformity index(CI) for planning target volume(PTV)and planning gross tumor volume(PGTV)between automatic and manual plans(P>0.05). The homogeneity index (HI) of PGTV in automatic and manual plans was 0.06 and 0.05, respectively(t=-6.92, P<0.05). Compared with manual plans, the automatic plans significantly decreased the V30 for urinary bladder by 2.7% and decreased the V20 for femoral head sand auxiliary structure(avoidance)by 8.37% and 15.95%, respectively (t=5.65, 11.24, P<0.05). Meanwhile, the average doses to bladder, femoral heads, and avoidance decreased by 1.91, 4.01, and 3.88 Gy, respectively(t=9.29, 2.80, 10.23, P<0.05) using the automatic plans. The time of automatic VMAT planning was (71.49±25.48)min in 20 cases. Conclusions The proposed automatic planning based on dose prediction and an iterative optimization algorithm is feasible and has great potential for sparing OARs and improving the utilization rate of clinical resources. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |