AccuContour软件在乳腺癌靶区和危及器官自动勾画中的研究  

Study on Automatic Segmentation of CTV and OARs by AccuContour Software in Breast Cancer

作　　者：龚筱钦 陈飞[2] 胡静[2] 游涛[2] 张开军[2] 戴春华[2] 朱雅群[1] GONG Xiaoqin;CHEN Fei;HU Jing;YOU Tao;ZHANG Kaijun;DAI Chunhua;ZHU Yaqun(Department of Radiation Oncology,The Second Affiliated Hospital of Soochow University,Suzhou Jiangsu 215000,China;Department of Radiation Oncology,Affiliated Hospital of Jiangsu University,Zhenjiang Jiangsu 212000,China)

机构地区：[1]苏州大学附属第二医院放疗科,江苏苏州215000 [2]江苏大学附属医院放疗科,江苏镇江212000

出　　处：《中国医疗设备》2022年第7期76-79,88,共5页China Medical Devices

基　　金：江苏省卫健委医学科研基金重点项目(ZDB2020022);镇江市重点研发计划(社会发展)项目(SH2021040)。

摘　　要：目的探讨基于U-net的AccuContour(AC)软件在乳腺癌临床靶区(Clinical Target Volume,CTV)和危及器官(Organs At Risk,OARs)自动勾画中的可行性。方法选取60例早期右侧乳腺癌保乳术后患者,由临床医生勾画CTV和OARs。随机抽取40例作为训练集,剩余20例作为测试集。对训练集进行学习形成新的自动勾画模型(Model-ST),利用戴斯相似系数(Dice Similarity Coefficient,DSC)、豪斯多夫距离(Hausdorff Distance,HD)及相对体积差值(Relative Volume Difference,RVD)分析其效果,并与AC软件自带的模型(Model-AC)比较。结果Model-ST中CTV的DSC值优于Model-AC(P<0.05)。OARs中,Model-AC双侧肺、肝脏、食管的DSC值均高于Model-ST,双肺的HD值以及左肺、气管和食管的RVD值均低于Model-ST,心脏的HD和RVD值高于Model-ST(P<0.05)。结论与Model-AC相比,Model-ST模型能够更准确地实现乳腺癌的靶区自动勾画,而Model-AC模型在危及器官自动勾画的准确性方面优于Model-ST。Objective To explore the feasibility of U-net-based AccuContour(AC)software in the automatic segmentation of clinical target volume(CTV)and organs at risks(OARs)in breast cancer.Methods A total of 60 cases of early breast cancer after breast conserving surgery were selected.The CTV and OAR swere manually delineated by physician.Of them,40 patients were randomly selected as the training set and the remaining 20 patients as the test set.Deep learning was carried out on the training set to set up an automatic segmentation mode(Model-ST).The sketching results were analyzed by using the dice similarity coefficient(DSC),Hausdorff distance(HD)and relative volume difference(RVD),and compared with the Model-AC provided by AC software.Results The DSC value of CTV in Model-ST was better than that of Model-AC(P<0.05).In the automatic segmentation comparison of OARs,the DSC values of bilateral lung,liver and esophagus of Model-AC were higher than those of Model-ST,the HD value of bilateral lungs and the RVD value of left lung,trachea and esophagus were lower than those of Model-ST,the HD and RVD value of heart is higher than Model-ST(P<0.05).Conclusion Compared with Model-AC,the Model-ST can achieve the automatic segmentation of CTV in breast cancer accurately,while the original Model-AC of AC software has better accuracy in automatic segmentation of OARs.

关 键 词：乳腺癌 自动勾画 深度学习 临床靶区 危及器官 

分 类 号：R815.2[医药卫生—放射医学]