基于卷积神经网络的纵隔淋巴结良恶性分类研究  

Study of mediastinal lymph node classification based on convolutional neural network

作　　者：程龙 秦航[1] 余晶[1] 蒋红兵 宋宁宁[1] CHENG Long;QIN Hang;YU Jing;JIANG Hongbing;SONG Ningning(Department of Medical Engineering,Nanjing First Hospital,Nanjing Medical University,Nanjing 210006,China;Nanjing Emergency Center,Nanjing 210003)

机构地区：[1]南京医科大学附属南京医院(南京市第一医院)临床医学工程处,南京210006 [2]南京市急救中心,南京210003

出　　处：《生物医学工程研究》2022年第4期365-368,共4页Journal Of Biomedical Engineering Research

基　　金：南京医科大学科技发展基金资助项目(NMUB2020213)。

摘　　要：为提高纵隔淋巴结良恶性分类的准确率及效率,本研究利用残差网络(residual network,ResNet)结合迁移学习方法,设计了一种纵膈淋巴结良恶性自动分类算法。实验结果显示,该方法的分类准确率、灵敏度、特异度可达75.1%、78.6%、84.4%,相比传统方法,分别提高了6%、9.9%、15.4%。本研究设计的多纵膈淋巴结自动分类算法相较于传统方法性能有显著提高,可作为临床辅助诊断的重要工具。In order to improve the accuracy and efficiency of benign and malignant classification of mediastinal lymph nodes,We used the residual network(ResNet)and combined with the transfer learning method to design an automatic classification algorithm for benign and malignant mediastinal lymph nodes.The experimental results showed that the classification accuracy,sensitivity and specificity of this method reached 75.1%,78.6%and 84.4%respectively,which were 6%,9.9%and 15.4%higher than that of the traditional methods.The performance of the automatic mediastinal lymph node classification algorithm designed in this study is significantly higher than that of traditional methods.It can be used as an important tool for clinical auxiliary diagnosis.

关 键 词：肺癌分期 淋巴结转移 支气管镜检查 人工智能 ResNet34 辅助诊断 

分 类 号：R318[医药卫生—生物医学工程] R445.1[医药卫生—基础医学] R581[自动化与计算机技术—控制理论与控制工程] TP181[自动化与计算机技术—控制科学与工程]