摘要
新冠肺炎(COVID-19)自爆发以来严重影响人类生命健康,近年来残差神经网络广泛应用于COVID-19识别任务中,辅助医生快速地诊断COVID-19患者,但是COVID-19图像病变区域形状复杂、大小不一,与周围组织的边界模糊,导致网络难以提取有效特征.本文针对上述问题,提出一种M^(3)Res-Transformer的新冠肺炎胸部X-ray图像识别模型,采用Res-Transformer作为模型的主干网络,结合ResNet和ViT,有效地整合局部病变特征和全局特征;设计混合残差注意力模块(mixed residual attention Module,mraM),同时考虑通道和空间位置的相互依赖性,增强网络的特征表达能力;为了增大感受野,提取多尺度特征,通过叠加具有不同扩张率的扩张卷积构造多尺度扩张残差模块(multiscale dilated residual Module,mdrM),根据不同层次特征尺度的差异,使用3个逐渐收缩尺度的mdrM进行多尺度特征提取;提出上下文交叉感知模块(contextual cross-awareness Module,ccaM),使用深层特征中的语义信息来引导浅层特征,然后将浅层特征中的空间信息嵌入深层特征中,采用交叉加权注意力机制高效聚合深层和浅层特征,获得更丰富的上下文信息.为了验证本文所提模型的有效性,在新冠肺炎胸部X-ray图像数据集上进行实验,与先进的CNN分类模型、融合不同注意力机制的ResNet50模型、基于Transformer的分类模型对比以及消融实验.结果表明,本文所提模型的Acc、Pre、Rec、F1-Score与Spe指标分别为96.33%、96.36%、96.33%、96.35%与96.26%,在COVID-19胸部X-ray图像识别任务中有效提升了识别精度,并通过可视化方法对其进行进一步验证,为COVID-19的辅助诊断提供重要的参考价值.
COVID-19 has seriously affected human life and health since its outbreak.In recent years,residual neural network has been widely used in COVID-19 recognition task to assist doctors to quickly diagnose COVID-19 patients.However,the shape of COVID-19 image lesion regions is complex,the size is different,and the boundary with surrounding tissues is blurred,which make it difficult for the network to extract effective features.Aiming at the above problems,a M^(3)Res-Transformer model for COVID-19 Chest X-ray image recognition is proposed.Res-Transformer is used as the back⁃bone network of the model,combining ResNet and ViT to effectively integrate local lesion features and global features;A mixed residual attention module(mraM)is designed to enhance the feature expression ability of the network by considering the interdependence of channels and spatial locations;In order to increase the receptive field and extract multi-scale fea⁃tures,the multi-scale dilated residual module(mdrM)is constructed by superimposing dilated convolution with different di⁃lation rates,and three mdrM with gradually shrinking scales are used for multi-scale feature extraction according to the dif⁃ference of feature scales at different layers;The contextual cross-awareness module(ccaM)is proposed,which uses the se⁃mantic information of deep features to guide shallow features,then embeds the spatial information of shallow features into deep features,and uses the cross-weighted attention mechanism to efficiently aggregate deep and shallow features to obtain richer contextual information.In order to verify the effectiveness of the model in this paper,experiments were conducted on the Chest X-ray image dataset of COVID-19,and through comparison with advanced CNN classification models,com⁃parison with ResNet50 models fusing different attention mechanisms,comparison with Transformer-based classification models and ablation experiment,the results showed that the Acc,Pre,Rec,F1-Score and Spe indexes of the proposed model are 96.33%,96.36%,96.33%,96.35%and 96.26%respectively,which effectively improves the recognition accuracy in CO⁃VID-19 Chest X-ray image recognition task,then it is further verified by visualization method,which provides important reference value for COVID-19 aided diagnosis.
作者
周涛
刘赟璨
侯森宝
常晓玉
叶鑫宇
陆惠玲
ZHOU Tao;LIU Yun-can;HOU Sen-bao;CHANG Xiao-yu;YE Xin-yu;LU Hui-ling(School of Computer Science and Engineering,North Minzu University,Yinchuan,Ningxia 750021,China;Key Laboratory of Image and Graphics Intelligent Processing of State Ethnic Affairs Commission,North Minzu University,Yinchuan,Ningxia 750021,China;School of Medical Information and Engineering,Ningxia Medical University,Yinchuan,Ningxia 750004,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2024年第2期589-601,共13页
Acta Electronica Sinica
基金
国家自然科学基金(No.62062003)
宁夏自治区重点研发计划(No.2020BEB04022)
北方民族大学研究生创新项目(No.YCX22198,No.YCX22190)。
