基于改进U-net的肺癌识别方法  

Lung cancer recognition method based on improved U-net

作　　者：张永梅[1] 彭炯 马健喆 胡蕾[3] ZHANG Yong-mei;PENG Jiong;MA Jian-zhe;HU Lei(School of Information Science and Technology,North China University of Technology,Beijing 100144,China;Department of Electronic and Information Engineering,The Hong Kong Polytechnic University,Hong Kong 999077,China;School of Computer Information Engineering,Jiangxi Normal University,Nanchang 330022,China)

机构地区：[1]北方工业大学信息学院,北京100144 [2]香港理工大学电子与信息工程系,香港999077 [3]江西师范大学计算机信息工程学院,江西南昌330022

出　　处：《计算机工程与设计》2021年第1期256-262,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61371143、61662033);教育部高等教育司产学合作协同育人基金项目(201801121002);全国高等学校计算机教育研究会2019年度课题基金项目(CERACU2019R05);“天诚汇智”创新促教基金项目(2018A03029);2019年度北京市教委基本科研业务费基金项目(110052971921/002)。

摘　　要：目前基于深度学习的肺癌辅助诊断方法存在无法准确定位病灶的缺陷。针对该问题,在现有U-net网络结构的基础上提出一种分两步走的基于改进U-net的肺癌识别方法。利用U-net获得病灶精确位置,通过CNN分类网络对病灶进行诊断,得到原始CT图像的检测结果。实验结果表明,该方法可以对肺部病灶进行较为精确的定位,分割效果的DSC相似度指数超过80%,对肺癌病灶进行分类诊断的准确率达到90.7%。The current lung cancer diagnosis methods based on deep learning cannot accurately locate lesions.To solve the problem,a two-step lung cancer recognition method based on improved U-net network structure was proposed.The location of the lesions was exacted through U-net network,the niduses were diagnosed through CNN classification network,and the lung can-cer detection results of the original CT image were provided.Experimental results show that the proposed method can effectively assist doctors in examining lung cancer since it can locate pulmonary niduses more accurately than the original U-net with the similarity index of DSC greater than 80%,and the classification accuracy rate of lung cancer is 90.7%.

关 键 词：肺结节 计算机断层扫描图像 U形网络 肺癌识别 CNN 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]