临床结直肠病理切片图像的自动辅助诊断  

Automatic auxiliary diagnosis of clinical histological images of colorectal pathology

作　　者：王莹[1] 段佳佳 祝闯 刘军[3] 董慧慧 齐勇刚 金木兰[1] WANG Ying;DUAN Jia-jia;ZHU Chuang;LIU Jun;DONG Hui-hui;QI Yong-gang;JIN Mu-lan(Department of Pathology,Capital Medical University Affiliated Beijing Chaoyang Hospital,Beijing 100020,China;Department of Invasive Technology,Coal General Hospital,Beijing 100028,China;Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]首都医科大学附属北京朝阳医院病理科,北京100020 [2]煤炭总医院介入科,北京100028 [3]北京邮电大学信息与通信工程学院,北京100876

出　　处：《临床与实验病理学杂志》2018年第10期1076-1079,共4页Chinese Journal of Clinical and Experimental Pathology

摘　　要：目的探讨借助深度学习算法进行结直肠病理组织切片的自动辅助诊断。方法收集首都医科大学附属朝阳医院病理科已确诊的92例增生性息肉、61例管状腺瘤、135例腺癌及100例神经内分泌肿瘤的存档病理切片,利用数字显微镜采集1 790张数字病理图像。其中1 530张图像作为训练集,260张图像作为测试集。利用深度神经网络基于训练集训练自动辅助诊断模型,并在测试集上进行测试。结果利用深度学习技术在结直肠病理图像测试集上的整体准确率≥91%,采用该方法对恶性肿瘤的灵敏度可达96. 7%。结论利用深度学习技术对结直肠病理组织切片的自动辅助诊断具有重要意义,不仅可以提高诊断效率,还能够降低漏诊风险。Purpose To study the automatic diagnosis of clinical colorectal pathology by using deep learning algorithm.Methods 92 cases of proliferative polyps,61 cases of tubular adenomas,135 cases of adenocarcinomas and 100 cases of neuroendocrine tumors were collected.These cases were diagnosed in the Department of Pathology,Beijing Chaoyang Hospital,Capital Medical University.1 790 digital pathological images were captured from the above pathological sections using digital microscope.Among the captured images,1 530 images were used as training set and 260 images were used as test set.An auto-assisted diagnosis model was trained based on the training set by using deep learning techniques,and a generated model was validated on the test set.Results The overall accuracy rate on the test set by using deep learning technology could reach over 91%.The sensitivity for malignant tumors of proposed method can reach 96.7%.Conclusion It is of great significance doing research on the automatic diagnosis of clinical colorectal pathology based on deep learning technology.It can not only improve the diagnosis efficiency,but also reduce the risk of misdiagnosis.

关 键 词：结直肠肿瘤 深度学习 病理切片 自动辅助诊断 

分 类 号：R445.9[医药卫生—影像医学与核医学] TP29[医药卫生—诊断学]