基于卷积神经网络的CT图像肺结节检测  

Detection of pulmonary nodules in CT images based on convolutional neural networks

作　　者：谢未央 陈彦博 王季勇 李强[1,4] 陈群 XIE Wei-yang;CHEN Yan-bo;WANG Ji-yong;LI Qiang;CHEN Qun(Research Center for Advanced Medical Imaging Technology,Shanghai Advanced Research Institute,Chinese Academy of Sciences,Shanghai 201210,China;School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China;School of Life Science and Technology,Shanghai Tech University,Shanghai 201210,China;Corporation Research Center,Shanghai United Imaging Healthcare Limited Company,Shanghai 201807,China)

机构地区：[1]中国科学院上海高等研究院高端医学影像技术研究中心,上海201210 [2]中国科学院大学电子电气与通信工程学院,北京100049 [3]上海科技大学生命科学与技术学院,上海201210 [4]上海联影医疗科技有限公司联影研究院,上海201807

出　　处：《计算机工程与设计》2019年第12期3575-3581,共7页Computer Engineering and Design

摘　　要：为帮助医生降低工作强度,减少诊断错误,提升准确率,提出一种基于三维卷积神经网络的肺结节检测算法。根据肺结节在CT图像中的特点,设计改进的三维候选区域推荐网络进行结节初始检测。在此基础上,使用多尺度、多网络融合的分类网络去除初检结果中的假阳性。在LUNA16数据集上验证了所提算法的准确性和有效性,并将结果与其它算法进行比较,讨论了该肺结节检测算法的性能。To help radiologists reduce work overload,improve accuracy and reduce misdiagnosis,apulmonary nodule detection algorithm based on three dimensional neural network was developed.Based on characteristics of pulmonary nodules in CT images,an improved three dimensional region proposal network was designed for initial nodule detection.Subsequently,a multiscale and ensemble convolutional neural network was employed to reduce the false positives.The accuracy and validity of the algorithm were assessed on LUNA16 dataset,and the results were evaluated and compared with other algorithms for performance discussion.

关 键 词：肺结节 CT图像 计算机辅助检测 卷积神经网络 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]