基于nnUnetv2算法的宫颈癌放疗患者直肠壁自动勾画效果评价  

Effect evaluation of rectum wall automatic delineation based on nnUnetv2 algorithm in cervical cancer patients received radiotherapy

作　　者：曾志鹏 左国平[1] ZENG Zhipeng;ZUO Guoping(School of Nuclear Science and Technology,University of South China,Hunan Hengyang 421001,China;Radiotherapy Center,Liuyang People's Hospital,Hunan Changsha 410300,China)

机构地区：[1]南华大学核科学技术学院,湖南衡阳421001 [2]浏阳市人民医院放疗中心,湖南长沙410300

出　　处：《现代肿瘤医学》2024年第11期2040-2045,共6页Journal of Modern Oncology

摘　　要：目的:基于人工智能算法研究直肠壁的自动勾画以及自动勾画直肠壁效果分析。方法:搜集2020至2023年在浏阳市人民医院接受放疗的191名宫颈癌患者的CT数据,由两位经验丰富的妇科医生进行直肠壁勾画,导出dicom数据及RT结构,运用nnUnetv2算法建立模型并进行训练,分析手动勾画与自动勾画的直肠壁戴斯系数和95%豪斯多夫距离,并且针对不同体积的直肠壁进行勾画效果分析。结果:自动勾画的整体直肠壁戴斯系数为0.85,内壁戴斯系数为0.92,外壁戴斯系数为0.94;整体直肠壁的95%豪斯多夫距离为3.03 mm,内壁为6.18 mm,外壁为4.99 mm,内外壁戴斯系数明显优于整体直肠壁系数,且外壁戴斯系数优于内壁,95%豪斯多夫距离整体直肠壁优于内外壁,对于有一定内容物(粪便或空气)的直肠壁,勾画效果比空直肠或者充盈的直肠要好;剂量学评估PTV的剂量参数,CI、HI以及直肠壁的V_(45)差异无统计学意义(P>0.05),但直肠壁的V_(40)及D_(mean)差异具有统计学意义(P<0.05),表明自动勾画的直肠壁对PTV的剂量参数无显著影响,但对器官本身的剂量参数有一定影响。结论:采用nnUnetv2算法进行直肠壁的自动勾画可以获得较好结果,能提升勾画效率及准确度,对计划剂量分布影响并不显著,可直接应用于临床计划设计。Objective:The study was based on the automatic delineation of the rectal wall and analysis of its effectiveness,utilizing artificial intelligence algorithms.Methods:CT data of 191 patients with cervical cancer who received radiotherapy in Liuyang People's Hospital from 2020 to 2023 were collected.Two experienced gynecologists outlined the rectal wall,derived dicom data and RT structure,and established a model with nnUnetv2 algorithm for training.The Dice coefficient and 95%Hausdorff distance of both manual and automatic rectum walls were analyzed,and the delineation effect was analyzed for different volume rectum walls.Results:The Dice coefficients of the whole rectum wall,inner wall and outer wall were 0.85,0.92 and 0.94 respectively.95%Hausdorff distance of the whole rectum wall was 3.03 mm,that of the inner wall was 6.18 mm,and that of the outer wall was 4.99 mm.The Dice coefficient of the inner and outer walls was significantly higher than that of the whole rectum wall,and the Dice coefficient of the outer wall was better than that of the inner wall.Meanwhile,95%Hausdorff distance to the overall rectum wall was better than the inter wall and outer walls.Rectal walls with some content(feces or air)were better delineated than empty or full rectum walls.There were no statistically significant differences in the dose parameters of PTV,CI,HI and V_(45) of the rectum wall in the dosimetric evaluation(P>0.05),but there were statistically significant differences in the V_(40)and D_(mean)of the rectum wall(P<0.05),indicated that the automatically delineated rectum wall had no significant impact on the dose parameters of PTV,but have a certain effect on the dose parameters of the organ itself.Conclusion:Using the nnUnetv2 algorithm to automatically sketch the rectal wall can obtain better results,improve the efficiency and accuracy of the sketch,and has no significant effect on the planned dose distribution,so it can be directly applied to clinical planning.

关 键 词：人工智能算法 自动勾画 直肠壁 戴斯系数 豪斯多夫距离 

分 类 号：R730.5[医药卫生—肿瘤]