Automatic Segmentation of Organs at Risk of Breast Cancer Using
U-Net-Based Automatic Segmentation Method
LI Hualing1
, LI Jinkai2
, ZHANG Wei3
, WANG Peipei2
, SUN Xinchen2
1. Department of Special Medicine, Nanjing Medical University, Nanjing Jiangsu 210009, China;
2. Department of Radiation Oncology, The First Affiliated Hospital with Nanjing Medical University, Nanjing Jiangsu 210009, China;
3. Manteia Data Technology Co., Ltd. in Xiamen Area Fujian Pilot Free Trade Zone, Xiamen Fujian 361000, China
Abstract:Objective To train and evaluate a U-Net-based automatic segmentation technique for delineating organs at risks (OARs)
of breast cancer. Methods A total of 140 cases of breast tumors were selected in the experiment, 120 of which were used as training
sets to build U-Net-based automatic segmentation model, and the remaining 20 cases were used as test sets to test the accuracy of the
automatic contouring methods. Dice similarity coefficient (DSC) and mean distance to agreement (MDA) were used to evaluate the
contouring results of the trained auto-segmentation model, and compared with the atlas-based auto-segmentation method with the
same training and test sets. Results The mean DSC values of OARs sketched by U-Net-based automatic contouring models were
higher than atlas-based automatic contouring methods; and the mean MDA values of OARs sketched by U-Net-based automatic
contouring models were lower than the latter. The differences in DSC values of the two auto-segmentation methods in bilateral lungs,
normal breasts, and heart, as well as the differences in MDA values in left lung, normal breasts, and heart were statistically significant
(P<0.05). Conclusion U-Net-based automatic segmentation model has a better contouring effect in OARs of breast cancer, and its
contouring accuracy is higher than that of atlas-based auto segmentation method.
李华玲1
,李金凯2
,张炜3
,王沛沛2
,孙新臣2. 基于U-Net的自动分割方法对乳腺癌危及器官的自动勾画[J]. 中国医疗设备, 2020, 35(8): 31-35.
LI Hualing1
, LI Jinkai2
, ZHANG Wei3
, WANG Peipei2
, SUN Xinchen2. Automatic Segmentation of Organs at Risk of Breast Cancer Using
U-Net-Based Automatic Segmentation Method. China Medical Devices, 2020, 35(8): 31-35.