基于U-Net的直肠癌肿瘤靶区和危及器官的自动分割模型

Automatic Segmentation Model of Rectal Cancer Gross Target Volume and Organs at Risk Based on U-Net

目的训练一种基于U-Net的自动分割模型用于直肠癌肿瘤靶区(GTV)和危及器官(OARs)的勾画,并评估该模型的勾画准确性及临床可行性。方法回顾性分析2018年1月至2020年10月于医院接受术前放射治疗的70例直肠癌患者的临床资料,随机分组,其中52例作为训练集,6例作为验证集,12例作为测试集。所勾画的结构为GTV和OARs,其中OARs包括左侧股骨头、右侧股骨头、膀胱,采用戴斯相似系数(DSC)及豪斯多夫距离(HD)评估勾画结果的准确性。结果GTV的平均DSC值为(0.70±0.10),平均HD值为(35.58±13.92)mm;左侧股骨头的平均DSC值为(0.94±0.04),平均HD值为(7.72±3.15)mm;右侧股骨头的平均DSC值为(0.93±0.02),平均HD值为(8.54±2.58)mm;膀胱的平均DSC值为(0.94±0.05),平均HD值为(9.42±4.21)mm。结论基于U-Net的自动分割模型在OARs的勾画中具有较高的准确性,但在GTV勾画中的准确性还有待提高,可将该模型用于辅助临床工作,以提高放射治疗医师的工作效率。

Objective An automatic segmentation model based on U-Net for the delineation of rectal cancer gross target volume(GTV)and organs at risk was trained and the accuracy of delineation and clinical feasibility of this model were evaluated.Methods The clinical data of 70 rectal cancer patients who underwent preoperative radiotherapy from January 2018 to October 2020 in the hospital was retrospectively analyzed.They were randomly divided into three groups,52 cases were defined as the training set,6 cases were defined as the verification set,and 12 cases were defined as the test set.The structure delineated were GTV and OARs(including the left femoral head,right femoral head and bladder).The Dice similarity coefficient(DSC)and Hausdorff distance(HD)were applied to evaluate the accuracy of the delineation results.Results The average DSC value of GTV was(0.70±0.10)and the average HD value was(35.58±13.92)mm;The average DSC value of the left femoral head was(0.94±0.04),and the average HD value was(7.72±3.15)mm;The average DSC value of the right femoral head was(0.93±0.02),and the average HD value was(8.54±2.58)mm;The average DSC value of the bladder was(0.94±0.05),and the average HD value was(9.42±4.21)mm.Conclusions The automatic segmentation model based on U-Net trained in this research has high accuracy in the delineation of OARs.Its accuracy in GTV delineation needs to be improved,but it can be used to assist clinical work to improve the work efficiency of radiotherapists.