摘要
目的:探讨基于Atlas实施宫颈癌危及器官自动勾画时图谱库入库病例数的增加对自动勾画结果的影响,以期得到最优图谱库病例数。方法:运用MIM软件建立4组宫颈癌图谱库(入库病例数目分别为30、60、90、120例)。随机选择图谱库外10例宫颈癌患者图像,由一名临床经验丰富的医生进行危及器官(膀胱、直肠和双侧股骨头)的手动勾画,将其定义为参考勾画,并对该10例患者图像进行危及器官自动勾画,勾画匹配数目分别选择为3和9。通过定量评价勾画时间、相似性系数(DSC)、敏感性指数(SI)、包容性指数、质心偏差、Jaccard系数(JAC)、Hausdorff距离(HD),将自动勾画结果与参考勾画进行单因素方差分析,从而探讨不同图谱库病例数对自动勾画结果的影响。结果:勾画匹配数目选择为3时,4组模板中平均自动勾画时间小于手动勾画(1.31/1.33/1.35/1.39min vs 10.25min),匹配数目选9时具有同样的趋势(5.07/5.24/5.14/5.24min vs 10.25min),但各组间没有差异性。匹配数目为3时膀胱SI(P=0.018)、直肠SI(P=0.010)、直肠DSC(P=0.016)、直肠JAC(P=0.013)、直肠HD(P=0.042),以及匹配数目为9时直肠HD(P=0.002)均具有统计学差异,其他参数没有统计学意义。结论:基于Atlas实施危及器官自动勾画能够节省勾画时间,模板数目的增加不会影响勾画效率,30例图谱库勾画时整体结果较差,60例以上的图谱库略有优势,提高膀胱、直肠的勾画准确性,但考虑时间成本,对于宫颈癌的勾画建议采用60例作为临床模板库病例数。
Objective To evaluate the effects of Atlas-based auto-segmentation (ABAS) with increasing Atlas database sizes on the automatic segmentation of organs-at-risk in patients with cervical cancer for obtaining the optimal Atlas size.Methods Four sets of cervical cancer Atlas databases,with 30,60,90 and 120 cases,respectively,were established with MIM software.The images of another 10 patients with cervical cancer were selected out of Atlas databases,and the organs-at-risk including bladder, rectum and bilateral femoral heads were segmented by an experienced radiation oncologist,and the obtained results were defined as reference volumes.Meanwhile,the 4 organs-at-risk of 10 patients were automatically segmented withABAS,with the matching number of 3 and 9,respectively.The time for segmentation,Dice similarity coefficient,sensitivity index,inclusiveness index, deviation of centroid,Jaccard coefficient and Hausdorff distance were quantitatively evaluated,and a one-way analysis of variance was performed on the results obtained with ABAS and reference contours,thereby discussing the effect of Altas database size on automatic segmentation.Results For the matching number of 3 and 9,the mean time of 4 sets of Atlas databases for automatic segmentation was shorter than that of manual segmentation (1.31/1.33/1.35/1.39 min vs 10.25 min;5.07/5.24/5.14/5.24 min vs 10.25 min),but there was no significant difference among different groups.When the matching number was 3,statistical differences were found in the sensitivity index of bladder and rectum (P=0.018,0.010),the Dice similarity coefficient,Jaccard coefficient and Hausdorff distance of rectum (P=0.016,0.013,0.042).When the matching number was 9,statistical difference was only found in the Hausdorff distance of rectum (P=0.002).The differences in the other parameters were trivial.Conclusion ABAS can shorten the time for segmentation,and the increasing of Atlas database size doesn't affect the segmentation efficiency. The Atlas database with 30 cases has a poor performance in se
作者
王金媛
徐寿平
刘博
郑庆增
张慧娟
杨微
曲宝林
WANG Jinyuan;XU Shouping;LIU Bo;ZHENG Qingzeng;ZHANG Huijuan;YANGWei;QU Baolin(Department of Radiotherapy,Chinese PLA General Hospital,Beijing 100853,China;Beihang Advanced Innovation Center for Big Date- based Precision Medicine,Beijing 100083,China;Image Processing Center,Beihang University,Beijing 100191,China;Department of Radiotherapy,Beijing Geriatric Hospital,Beijing 100095,China)
出处
《中国医学物理学杂志》
CSCD
2019年第7期760-764,共5页
Chinese Journal of Medical Physics
基金
国家重点研发计划(2017YFC0112100)
解放军总医院临床科研扶持基金(2017FC-TSYS-3005)
