增强CT中基于深度神经网络的卵巢恶性肿瘤转移性淋巴结术前辅助诊断系统的研究  

Clinical application of preoperative aided diagnosis system for metastatic lymph nodes of pelvic and abdominal cavity based on deep neural network in enhanced CT

作　　者：刘伟[1] 张丹[1] 宋克娟[2] 吕腾[2] 卢云[2] 姚勤[2] Liu Wei;Zhang Dan;Song Kejuan(Qingdao University,Qingdao 266071;Affliated Hospital of Qingdao University,Qingdao 266555)

机构地区：[1]青岛大学,青岛266071 [2]青岛大学附属医院,青岛266555

出　　处：《现代妇产科进展》2020年第10期726-729,733,共5页Progress in Obstetrics and Gynecology

摘　　要：目的:探讨在增强CT中基于深度神经网络建立的卵巢恶性肿瘤盆腹腔转移性淋巴结辅助诊断系统在临床应用中的价值。方法:收集2016年1月至2019年6月于青岛大学附属医院妇产科进行治疗的卵巢恶性肿瘤患者的薄层增强CT影像,并对其转移性淋巴结进行标记,随机分为实验组及验证组。应用Faster R-CNN算法,对实验组图像进行学习,建立转移性淋巴结辅助诊断系统,并用验证组图像进行验证。采用受试者工作特征曲线对辅助诊断系统进行评估。结果:通过对实验组薄层增强CT图像的学习,初步建立了卵巢恶性肿瘤转移性淋巴结自动诊断系统,在不断的迭代训练中人工智能的损失函数值不断降低,人工智能诊断相应的受试者工作特征曲线下面积为0.7664。结论:通过测试结果来看,基于Faster R-CNN的卵巢恶性肿瘤转移性淋巴结术前辅助诊断系统的性能已达到较高水平。虽与影像科医师的诊断水平尚有差距,但随着训练样本的增加及算法的改进,还有进一步提高的空间,对于术前辅助诊断转移性淋巴结,以及进行更有效的淋巴结清扫,具有较高的临床应用价值。Objective:To explore the value of the auxiliary diagnostic system of pelvic and abdominal metastatic lymph nodes of ovarian malignant tumors based on deep neural networks in enhanced CT.Methods:Retrospectively collected enhanced CT images of patients with ovarian malignancies who were treated in the obstetrics and gynecology department of Qingdao University Hospital from January 2016 to June 2019.After metastatic lymph nodes were labeled,they were randomly divided into experimental groups and verified group.The Faster R-CNN algorithm was used to study the experimental group images,and then a metastatic lymph node assisted diagnosis system was established,and the validation group images were used for verification.The receiver operating characteristic curve was used to evaluate the auxiliary diagnosis system.Result:Through the study of enhanced CT images of the experimental group,an automatic diagnostic system for ovarian malignant metastatic lymph nodes was initially established.During the continuous iterative training,the value of the loss function of artificial intelligence continued to decrease,and the area under the corresponding ROC curve of artificial intelligence diagnosis was 0.7664.Conclusions:According to the test results,the diagnostic level of metastatic lymph nodes of ovarian tumors based on Faster R-CNN has reached a high level.Although it has not reached the level of radiologist,with the increase of training images and algorithms,there is still room for further improvement.It has high clinical application value for the diagnosis of metastatic lymph nodes before surgery and for more effective and effective lymph node dissection.

关 键 词：人工智能 增强CT 卵巢恶性肿瘤 盆腹腔 转移性淋巴结 

分 类 号：R711[医药卫生—妇产科学]