AccuContour软件自动勾画镜面人患者4D-CT图像心脏结构的应用分析

Application Analysis of the AccuContour Software in Automatic Delineating the Structure of Heart with the 4D-CT images for Mirror People

目的评估AccuContour软件自动勾画镜面人患者四维CT(4D-CT)图像心脏结构的可行性与准确性。方法选取1例镜面人肺癌患者的10套4D-CT图像序列进行研究,10套图像序列对应10个呼吸时相。对10套CT序列,使用Eclipse治疗计划系统手工勾画心脏结构;使用AccuContour软件自动勾画心脏结构;使用WPS表格,对自动勾画和手工勾画的体积差异、位置差异、形状一致性和勾画时间进行统计分析。结果自动和手工勾画两种方式心脏结构的体积差异(△V%)为(5.63±2.80)%;总位置差异(ΔL)为(2.28±0.45)mm;Dice相似性系数、敏感性指数、包容性指数和Jaccard系数的值分别为0.909±0.011、0.936±0.012、0.884±0.020和0.835±0.019;自动勾画时间平均约为49 s。结论使用基于深度学习的AccuContour软件自动勾画镜面人患者4D-CT图像的心脏结构,能够达到比较满意的勾画结果,使用该软件可明显缩短勾画时间,有效提高工作效率。

Objective To evaluate the feasibility and accuracy of the AccuContour software in automatic delineating the heart structure with 4D-CT images for mirror people.Methods Ten sets of 4D-CT image series of mirror people with lung cancer were studied.Ten sets of CT series were corresponding to ten respiratory phases.For the ten sets of CT series,the heart structures were manually delineated in the treatment planning system of Eclipse,and automatically delineated by the AccuContour software.The WPS sheet was used to statistically analyze the differences in volume,position,shape,and delineation time between the automatic and manual delineations.Results The differences in volume(ΔV%)between the two methods was(5.63±2.80)%;the total position difference(ΔL)was(2.28±0.45)mm;the values of dice similarity coefficient,sensitivity index,inclusiveness index and Jaccard coefficient were 0.909±0.011,0.936±0.012,0.884±0.020 and 0.835±0.019,respectively;the average time of automatic delineation was about 49 s.Conclusion The AccuContour software based on deep learning could achieve satisfactory results of automatic delineation of the heart structure of 4D-CT images for the mirror people.By using the software,it could significantly shorten the delineation time and effectively improve the work efficiency.