卷积神经网络在T3/4期胃癌影像学诊断中应用  

APPLICATION OF CONVOLUTIONAL NEURAL NETWORK IN RADIOLOGICAL DIAGNOSIS OF T3/4 GASTRIC CANCER

作　　者：张训营 张凯明 张超[1] 马金龙[2] 卢云[1,3] 王东升 ZHANG Xunying;ZHANG Kaiming;ZHANG Chao;MA Jinlong;LU Yun;WANG Dongsheng(Department of Gastrointestinal Surgery,The Affiliated Hospital of Qingdao University,Qingdao 266003,China)

机构地区：[1]青岛大学附属医院胃肠外科,山东青岛266003 [2]青岛大学附属医院影像科 [3]山东省数字医学与计算机辅助手术重点实验室

出　　处：《青岛大学学报（医学版）》2021年第5期731-735,共5页Journal of Qingdao University(Medical Sciences)

基　　金：青岛大学附属医院青年科研基金项目(3458)。

摘　　要：目的利用卷积神经网络构建T3/4期胃癌自动识别平台以达到辅助临床诊疗目的。方法回顾性收集208例胃癌病人的增强CT图像,并按照7∶1比例随机分入训练集(182例)和验证集(26例),利用labelImg软件标识病变区域,用训练集对平台进行训练,用验证集进行验证。通过对比平台和影像学专家标识图像信息,采用受试者工作特征(ROC)曲线,对平台性能进行评估,评价指标包括ROC曲线下面积(AUC)、准确度、灵敏度、特异度、阳性预测值及阴性预测值等。结果平台的AUC为0.924,对T3/4期胃癌识别的准确度、灵敏度、特异度分别为0.927、0.924、0.930,阳性预测值为0.933,阴性预测值为0.921。结论平台基于增强CT对T3/4期胃癌的识别准确性与高年资影像学专家相当,并可以准确识别出T3/4期胃癌病变区域,极大提高了胃癌术前诊断效率。Objective To establish an automatic recognition platform for T3/4 gastric cancer based on convolutional neural network, and to achieve the purpose of assisting clinical diagnosis and treatment. Methods A retrospective analysis was performed for the contrast-enhanced CT images of 208 patients with gastric cancer, and the patients were randomly divided into trai-ning set with 182 patients and validation set with 26 patients at a ratio of 7∶1. The labelImg software was used to identify the lesion area, and the platform was trained by the training set and validated by the validation set. By comparing the image information identified by the platform and imaging experts, the receiver operating characteristic(ROC) curve was used to evaluate the perfor-mance of this platform, and assessment indices included the area under the ROC curve(AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. Results The platform had an AUC of 0.924, with an accuracy of 0.927, a sensitivity of 0.924, and a specificity of 0.930 in identifying T3/4 gastric cancer, and the platform had a positive predictive value of 0.933 and a negative predictive value of 0.921. Conclusion The platform based on contrast-enhanced CT has a comparable accuracy to senior imaging experts in identifying T3/4 gastric cancer and can accurately identify the area of the lesion of T3/4 gastric cancer, which greatly improves the efficiency of preoperative diagnosis of gastric cancer.

关 键 词：卷积神经网络 胃肿瘤 肿瘤分期 人工智能 诊断 

分 类 号：R735.2[医药卫生—肿瘤] R445.3[医药卫生—临床医学]