胸部X光片多特征融合的冠状病毒诊断  

Multi-feature fusion of chest X-ray for coronavirus diagnosis

作　　者：龙如山 杨丹[1] LONG Rushan;YANG Dan(School of Computer Science and Software Enginering,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]辽宁科技大学计算机与软件工程学院,辽宁鞍山114051

出　　处：《辽宁科技大学学报》2023年第1期58-64,共7页Journal of University of Science and Technology Liaoning

基　　金：辽宁省教育厅高校基本科研项目(LJKMZ20220646)。

摘　　要：针对目前网络模型从X光片提取的特征过于单一,并且监督学习网络太过依赖疾病诊断标签,得到的X光片特征不利于冠状病毒诊断。提出一种胸部X光片多特征融合的冠状病毒诊断框架Covid-19Net。使用EfficientNet网络作为提取X光片特征的主干网络,采用教师—学生对比学习网络优化X光片特征,利用X光片肺分割生成的肺掩膜图像特征和X光片中提取关键点特征,聚合X光片提取的多种特征进行冠状病毒诊断。实验结果表明,Covid-19Net方法可以提高冠状病毒诊断性能。The features extracted from X-ray films by the current network model are too simplistic,the supervised learning network relies too much on disease diagnosis labels,and the obtained X-ray film features are not conducive to the diagnosis of coronaviruses.A coronavirus diagnostic framework Covid-19Net based on the multi-feature fusion of chest X-rays is proposed.First,the EfficientNet network is used as the backbone network for extracting X-ray film features,and then the teacher-student contrastive learning network is used to optimize the X-ray film features,and then the lung mask image features generated by X-ray film lung segmentation and key points are extracted from the X-ray film feature.Finally,multiple features extracted from the aggregated X-ray film are used for coronavirus diagnosis.The experimental results show that the Covid-19Net method can improve the performance of coronavirus diagnosis.

关 键 词：新冠病毒 对比学习 多特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]