基于改进EfficientNet网络的肺结节图像分类研究
Research on image classification of pulmonary nodules based on improved EfficientNet network
周孟然 1王宁 2高立鹏 1王昊男 1卞凯 1刘思怡1
作者信息
- 1. 安徽理工大学电气与信息工程学院,安徽淮南 232001
- 2. 安徽理工大学力学与光电物理学院,安徽淮南 232000
- 折叠
摘要
针对目前计算机辅助诊断肺结节良恶性精度值较低、误诊率较高以及模型较复杂等问题,提出一种改进EfficientNet网络的肺结节良恶性分类模型.首先,在特征提取部分融合ECA模块,搭建出EMBConv结构,使网络模型关注更多特征信息;其次,使用跨域迁移学习,提高了网络模型的分类性能;然后采用Ranger优化器优化网络训练,有效防止模型陷入局部最优;最后,将从LIDC-IDRI数据集分割提取的肺结节图像输入到改进的分类模型中.实验结果表明,所提方法在网络参数量和计算量表现出较强竞争力,同时,分类准确率、精确率达到了91.83％和95.50％,较模型改进前分别提高了1.66％和4.41％.
Abstract
Aiming at the current computer-aided diagnosis of benign and malignant pulmonary nodules with low accuracy,high misdiagnosis rate and complex model,an improved EfficientNet network classification model for benign and malignant pulmonary nodules was proposed.First,the EC A module is integrated in the feature extraction part,and the EMBConv structure is built to make the network model focus on more feature information.Secondly,the cross-domain transfer learning is used to improve the classification of the network model performance.Then the Ranger optimizer is used to optimize the network training to effectively prevent the model from falling into local optimum.Finally,the lung nodule images extracted from the LIDC-IDRI dataset are input into the improved classification model.The experimental results show that the proposed method shows strong competitiveness in the amount of network parameters and calculations,and the classification accuracy and precision have reached 91.83％and 95.50％,respectively,compared with the model before the improvement 1.66％and 4.41％.
关键词
肺结节分类/ECA模块/跨域迁移学习/Ranger优化器Key words
pulmonary nodule classification/ECA module/cross-domain transfer learning/Ranger optimizer引用本文复制引用
基金项目
安徽省科技重大专项项目(201903a07020013)
教育部产学研创新基金(2019ITA01010)
安徽理工大学研究生创新基金项目(2022CX1007)
出版年
2024
NETL
NSTL
万方数据