基于深度学习与图谱库两种自动勾画方法对直肠癌危及器官勾画效果的对比研究  

Effects of deep learning-versus atlas-based automatic contouring methods on the contouring of organs-at-risk in rectal cancer

作　　者：李玉成 王成 贾勇士[1] 唐建明 詹文明[1] 李强[1] 邱凌云[1] 陈维军 Li Yucheng;Wang Cheng;Jia Yongshi;Tang Jianming;Zhan Wenming;Li Qiang;Qiu Lingyun;Chen Weijun(Radiotherapy Center,Zhejiang Provincial People's Hospital(Affiliated Hospital of Hangzhou Medical College),Hangzhou 310014,Zhejiang Province,China;School of Nuclear Science and Technology,University of South China,Hengyang 421000,Hunan Province,China)

机构地区：[1]浙江省人民医院杭州医学院附属人民医院放疗中心,310014 [2]南华大学核科学技术学院,湖南省衡阳421000

出　　处：《中国基层医药》2021年第10期1490-1495,共6页Chinese Journal of Primary Medicine and Pharmacy

基　　金：浙江省基础公益研究计划项目(GF21H180053);浙江省医药卫生科技计划项目(2021PY002)。

摘　　要：目的比较基于深度学习的AiContour®©与图谱库的Raystation®©两种自动勾画方法对放射治疗中直肠癌患者CT图像的危及器官勾画效果,为临床应用提供依据。方法选取浙江省人民医院2020年1-6月治疗的直肠癌患者50例,资深放疗医生勾画好的直肠癌患者20例的CT图像作为目标图像,分别基于两种软件的数据模板库对其进行自动勾画。采用豪斯多夫距离指数(HD指数)、平均距离指数(MDA指数)、形状相似性指数(DSC指数)、贾卡德指数(JC指数)四种指标定量评估两种自动勾画方法获得的危及器官轮廓体积的精确性。结果两种勾画方法除左股骨头[(6.81±2.66)与(7.24±2.10)]、右股骨头[(7.38±3.91)与(8.14±3.71)]、骨盆[(24.00±9.01)与(24.66±9.67)]的HD指数差异无统计学意义(t左股骨头=-0.831,t右股骨头=-0.821,t骨盆=-0.357,均P>0.05)外,其他危及器官的勾画参数差异均有统计学意义(均P<0.05)。AiContour®©自动勾画的DSC指数的均值均>0.7,而Raystation®©自动勾画的DSC值除了左肾、右肾、直肠、膀胱DSC值<0.7,其他危及器官DSC值均>0.7。此外,AiContour®©软件勾画出来的HD、MDA、JC值普遍优于Raystation®©勾画的结果。结论两种自动勾画软件勾画的危及器官经过略微修改均能满足临床使用,但是AiContour®©勾画效果要优于Raystation®©勾画的效果。Objective To investigate the effects of deep learning-based AiContour®©versus atlas-based Raystation®©automatic contouring methods on the contouring of organs-at-risk on CT images of patients with rectal cancer who undergo radiotherapy,providing evidence for clinical application.Methods Fifty patients with rectal cancer who received treatment during January to June 2020 in Zhejiang Provincial People's Hospital(Affiliated Hospital of Hangzhou Medical College)were included in this study.The CT images from 20 patients with rectal cancer that had been contoured by experienced radiotherapist were selected as target images and automatically contoured using the data template library of AiContour®©and Raystation®©automatic contouring methods.Hausdorff distance,mean distance to agreement,dice similarity coefficient,Jaccard coefficient were used to quantitatively evaluate the accuracy of the volume of contour of organs-at-risk automatically sketched by the two methods.Results There was no significant difference in Hausdorff distance in left femoral head[(6.81±2.66)vs.(7.24±2.10)],right femoral head[(7.38±3.91)vs.(8.14±3.71)],pelvis[(24.00±9.01)vs.(24.66±9.67)]between AiContour®©and Raystation®©automatic contouring methods(tleft femoral head=-0.831,tright femoral head=-0.821,tpelvis=-0.357,all P>0.05).Significant differences were observed in mean distance to agreement,dice similarity coefficient and Jaccard coefficient of organs-at-risk(all P<0.05).The mean values of dice similarity coefficient automatically sketched by AiContour®©method were>0.7.The DSC of left kidney,right kidney,rectum and bladder automatically sketched by Raystation®©method were<0.7,and the dice similarity coefficient values of other organs-at-risk automatically sketched by Raystation®©method were>0.7.In addition,Hausdorff distance,mean distance to agreement and Jaccard coefficient values of organs-at-risk automatically sketched by AiContour®©method were superior to those automatically sketched by Raystation®©.Conclusion Afte

关 键 词：直肠肿瘤 放射疗法 危及器官 深度学习 自动勾画 豪斯多夫指数 平均距离指数 贾卡德指数 形状相似性指数 

分 类 号：R735.37[医药卫生—肿瘤]