摘要
为解决现有肺结节检测模型精度低、漏诊率和误诊率高等问题,提出融合残差模块的肺结节检测算法。候选结节检测阶段,提出残差U-Net(residual U-Net,RU-Net)分割网络,将改进的残差网络(residual network,ResNet)模块与U-Net结构融合,提升模型特征提取能力;加入改进的损失函数解决数据类别不均衡问题,提高检测敏感度。假阳性减少阶段,采用三维卷积神经网络(3D CNN)用于候选结节分类,充分获得结节空间信息,达到降低假阳性的目的。实验结果表明,该算法能够准确并高效地分割和检测肺结节。
To solve the problems of low accuracy,high rate of missed diagnosis and high rate of misdiagnosis of existing lung no-dule detection models,a lung nodule detection algorithm combining a residual module was proposed.In the candidate nodule detection stage,the residual U-Net(RU-Net)segmentation network was proposed.The improved residual network(ResNet)module was integrated with the U-Net structure to enhance the model feature extraction ability.The improved loss function solved the problem of imbalanced data categories to obtain higher detection sensitivity.In the false positive reduction stage,a three-dimensional convolutional neural network(3D CNN)was used for candidate nodule classification,which fully obtained nodule spatial information,and achieved the purpose of reducing false positives.Experimental results show that the proposed algorithm can accurately and efficiently segment and detect lung nodules.
作者
马巧梅
梁昊然
郎雅琨
MA Qiao-mei;LIANG Hao-ran;LANG Ya-kun(Software School,North University of China,Taiyuan 030051,China;Shanxi Military and Civilian Integration Software Technology Engineering Research Center,Taiyuan 030051,China)
出处
《计算机工程与设计》
北大核心
2021年第4期1058-1064,共7页
Computer Engineering and Design
基金
山西省自然科学基金项目(201801D121026)
山西省研究生创新基金项目(2020SY398)。
关键词
计算机辅助诊断
CT图像
肺结节
深度学习
卷积神经网络
残差网络
computer-aided diagnosis
computer tomography image
lung nodule
deep learning
convolutional neural network
residual network
