人工智能云技术在乳腺癌患者心脏亚结构自动勾画中的应用

目的:评估人工智能云技术勾画平台（AI Contour）在乳腺癌患者心脏亚结构自动勾画中的准确性和可行性。方法:选取10例进行乳腺癌放射治疗患者的血管增强CT作为研究对象。在AI Contour上分别采用手动勾画、自动勾画和自动勾画后手动修改模式来完成10例患者的心脏亚结构勾画，包括左心房、右心房、左心室、右心室。比较Dice相似性系数（DSC）、Jaccard系数（JC）、Hausdorf距离（HD）、质心偏差（CMD）、包容性系数（IncI）、敏感性指数（SI）、勾画时间。结果:以手动勾画为金标准，自动勾画与手动勾画各心脏亚结构的DSC>0.8，JC>0.6，HD<9 mm，CMD<5 mm，IncI>0.8，SI>0.7。自动勾画后手动修改进一步提高了勾画精度，其中JC>0.8。自动勾画时间与手动勾画时间为（85.50±6.06） s vs （1 160.30±74.31） s，差异具有统计学意义（P<0.05）。自动勾画后手动修改总时间与手动勾画时间为（558.70±33.40） s vs （1 160.30±74.31） s，差异具有统计学意义（P<0.05）。结论:通过比较发现自动勾画技术能以较高的精度完成乳腺癌患者左心房、右心房、左心室、右心室的勾画，节省了大量时间，自动勾画后手动修改能进一步提高各心脏亚结构的勾画精度，同时云勾画平台具有远程协作的优势，值得推广运用。

Application of artificial intelligence cloud technology in auto-segmentation of cardiac substructure of breast cancer patients

Abstract: Objective To evaluate the accuracy and feasibility of artificial intelligence cloud technology contouring platform (AI Contour) in auto-segmentation of cardiac substructure of breast cancer patients. Methods The vascular enhanced CTs from 10 patients underwent breast cancer radiotherapy from July 2019 to December 2019 were selected as the research object. On AI Contour, manual segmentation, auto-segmentation and manual modification after auto-segmentation were used to segment the cardiac substructures of 10 patients, including left atrium, right atrium, left ventricle and right ventricle. The differences of data in Dice similarity coefficient (DSC), Jaccard coefficient (JC), Hausdorf distance (HD), Center of Mass Deviation (CMD), inclusiveness coefficient (IncI), sensitivity index (SI), and segmentation time were compared. Results With manual segmentation as the gold standard, the results of each cardiac substructure made by auto-segmentation and manual segmentation have differences of DSC>0.8, JC>0.6, HD<9 mm, CMD<5 mm, IncI>0.8, SI>0.7. Manual modification after auto-segmentation further improved the segmentation accuracy, in which JC>0.8. The time of auto-segmentation and manual segmentation were (85.50±6.06) s vs. (1160.30±74.31) s respectively, and the difference was statistically significant (P<0.05). The total time of manual modification after auto-segmentation and manual segmentation were (558.70±33.40) s vs. (1160.30±74.31) s, and the difference was statistically significant (P<0.05). Conclusions Through comparison, it was found that the auto-segmentation technique can segment the left atrium, right atrium, left ventricle, and right ventricle of breast cancer patients with high accuracy, saving a lot of time. Manual modification after auto-segmentation can further improve the segmentation accuracy of cardiac substructure of each part. The cloud segmentation platform has the advantage of remote collaboration, and is worth popularizing.