摘要
针对肺部X射线图像的病灶区域较小、形状复杂,与正常组织间的边界模糊,使得肺炎图像中的病灶特征提取不充分的问题,提出了一个面向特征增强的双残差Res-Transformer肺炎识别模型,设计3种不同的特征增强策略对模型特征提取能力进行增强。设计了组注意力双残差模块(GADRM),采用双残差结构进行高效的特征融合,将双残差结构与通道混洗、通道注意力、空间注意力结合,增强模型对于病灶区域特征的提取能力;在网络的高层采用全局局部特征提取模块(GLFEM),结合CNN和Transformer的优势使网络充分提取图像的全局和局部特征,获得高层语义信息的全局特征,进一步增强网络的语义特征提取能力;设计了跨层双注意力特征融合模块(CDAFFM),融合浅层网络的空间信息以及深层网络的通道信息,对网络提取到的跨层特征进行增强。为了验证本文模型的有效性,分别在COVID-19 CHEST X-RAY数据集上进行消融实验和对比实验。实验结果表明,本文所提出网络的准确率、精确率、召回率,F1值和AUC值分别为98.41%,94.42%,94.20%,94.26%和99.65%。DRT Net能够帮助放射科医生使用胸部X光片对肺炎进行诊断,具有重要的临床作用。
Deep learning for lung X-ray image recognition has emerged as a prominent research area.The challenge lies in the small,complexly shaped lesion areas within lung X-rays,where the boundary between the lesion and normal tissue is often unclear,complicating feature extraction in pneumonia images.This paper introduces a Dual Res-Transformer pneumonia recognition model focused on feature enhance-ment.It incorporates three feature enhancement strategies to augment the model's feature extraction capa-bilities.The model's key components include:the Group Attention Dual Residual Module(GADRM),which leverages a dual-residual structure for effective feature fusion and enhances local feature extraction through channel shuffle,channel attention,and spatial attention;the Global-Local Feature Extraction Module(GLFEM),which applies at the network's higher levels,merging CNN and Transformer benefits to extract comprehensive global and local image features,thereby boosting the network's semantic feature extraction;and the Cross-layer Dual Attention Feature Fusion Module(CDAFFM),designed to merge shallow network spatial information with deep network channel information,enhancing the network's cross-layer feature extraction.The model's efficacy was validated through ablation and comparative experi-ments on the COVID-19 CHEST X-RAY dataset.Results demonstrate the network's high performance,with accuracy,precision,recall,F1 score,and AUC values of 98.41%,94.42%,94.20%,94.26%,and 99.65%,respectively.This model offers significant assistance to radiologists in diagnosing various pneu-monia cases using chest X-rays,marking a crucial advancement in computer-aided pneumonia diagnosis.
作者
周涛
彭彩月
杜玉虎
党培
刘凤珍
陆惠玲
ZHOU Tao;PENG Caiyue;DU Yuhu;DANG Pei;LIU Fengzhen;LU Huiling(College of Computer Science and Engineering,North Minzu University,Yinchuan 750021,China;Key Laboratory of Image and Graphics Intelligent Processing of State Ethnic Affairs Commission,North Minzu University,Yinchuan 750021,China;School of Medical Information&Engineering,Ningxia Medical University,Yinchuan 750004,China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2024年第5期714-726,共13页
Optics and Precision Engineering
基金
国家自然科学基金资助项目(No.62062003)
宁夏自然科学基金资助项目(No.2022AAC03149)。
