摘要
目前,肺癌的发病率和致死率高居癌症首位,对其进行早期诊治对于提高患者的生存率和改善预后极其重要。肺结节是肺癌的早期表现,临床上医生通过观察对其分割后的体积和形态等特征来进行良恶性诊断,然而采用人工的方式进行肺结节分割非常低效。本文提出了一种基于MSVNet网络的肺结节分割方法,该网络继承了原始VNet的结构,同时引入了多尺度特征结构,通过提取肺结节图像的多尺度特征,同时利用深监督策略进行特征优化,能够有效地提升模型的分割性能。本文利用LIDC-IDRI肺结节公开数据集对模型的性能进行了评估,结果表明,本文方法所取得的分割结果与金标准相近,具有良好的肺结节分割性能,以及较高的分割鲁棒性,对不同大小的肺结节均能取得较好的分割效果。
At present,the incidence and mortality of lung cancer are the highest among cancers.Early diagnosis and treatment of lung cancer are extremely important to improve the survival rate and prognosis of patients.Pulmonary nodules are the early manifestation of lung cancer,which are clinically diagnosed as benign or malignant by observing the characteristics such as volume and morphology after segmentation.However,manual segmentation of pulmonary nodules is very inefficient.In this study,a segmentation method of pulmonary nodules based on MSVNet network is proposed,which inherits the structure of the original VNet,meanwhile a multi-scale feature structure is introduced.Through extracting the multi-scale feature of pulmonary nodule image and optimizing the feature with deep supervision strategy,the segmentation performance of the model can be effectively improved.In this study,the performance of the model is evaluated using the LIDC-IDRI pulmonary nodule public data set.The results show that the segmentation results obtained with the proposed method are similar to the gold standard.The proposed method has good pulmonary nodule segmentation performance and high segmentation robustness,and can achieve good segmentation results for the pulmonary nodules with different sizes.
作者
钟思华
王梦璐
郭兴明
张瑶
郑伊能
Zhong Sihua;Wang Menglu;Guo Xingming;Zhang Yao;Zheng Yineng(Chongqing Engineering Research Center for Medical Electronics Technology,College of Biological Engineering,Chongqing University,Chongqing 400044,China;Corporation Research Center,Shanghai United Imaging Healthcare Co.,Ltd.,Shanghai 201807,China;Department of Radiology,The First Affiliated Hospital of Chongqing Medical University,Chongqing 400044,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2020年第9期206-215,共10页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(31570003)项目资助
关键词
肺结节分割
深度学习
多尺度特征
深监督
pulmonary nodule segmentation
deep learning
multi-scale feature
deep supervision
作者简介
钟思华,分别在2017年、2020年于重庆大学获得学士、硕士学位,现为上海联影医疗科技有限公司科研合作专家,主要研究方向为医学图像处理。E-mail:zhongsh@cqu.edu.cn;通信作者:郭兴明,分别在1984年、1991年和1994年于重庆大学获得学士、硕士和博士学位,现为重庆大学教授、博士生导师,主要研究方向为生物医学信号检测及仪器、远程医疗及医学图像处理。E-mail:guoxm@cqu.edu.cn
