Abstract：Objective To analyze and compare the effects of the auto-segmentation method based on deep learning (DL) and Atlas for
contouring organs-at-risk (OARs) in radiotherapy for tumors locating in upper abdomen. Methods The CT images of 27 patients with
tumors in abdomen were selected and automatically segmented by DL and Atlas-based methods. By comparing the OARs of manual
contouring with those of automatic segmentation, the Hausdorff distance, mean distance to agreement, dice similarity coefficient
（DSC）and Jaccard similarity coefficient（Jaccard）of the two contouring results were calculated. By comparing the above four
parameters, we can evaluated the accuracy of the two methods in segmenting the upper abdominal OARs. Results DL method was
superior to Atlas in the four evaluation parameters of left and right kidney outline, and the differences were statistically significant
(P<0.05). For liver segmentation, DL method was also superior to Atlas method in the evaluation of Jaccard similarity coefficient
and the difference was statistically significant (P=0.03). For spinal cord segmentation, both DSC and Jaccard showed that DL was
inferior to Atlas (P<0.05). Conclusion DL method has a certain advantage in judging the left and right kidneys and livers relative to
Atlas method for the automatic segmentation of OARs in upper abdomen. It can obtain more accurate outline results, so as to better
assist radiotherapy doctors to complete the outline work of OARS.
高山宝，侯震，李双双，刘娟，闫婧. 两种自动勾画方法对上腹部危及器官勾画结果对比分析[J]. 中国医疗设备, 2021, 36(6): 75-78.
GAO Shanbao, HOU Zhen, LI Shuangshuang, LIU Juan, YAN Jing. Comparison and Analysis of Two Auto-Segmentation Methods for the
Upper Abdominal Organs-at-Risk. China Medical Devices, 2021, 36(6): 75-78.
