融合注意力机制与集成学习的肺炎CT图像智能诊断研究

打开文本图片集

中图分类号：TP319.4 文献标识码：A文章编号：1006-8228（2025）12-19-07

Research on Intelligent Diagnosis Pneumonia CT Images Integrating Attention Mechanisms and Ensemble Learning

DiaoLijuan，Wu Wei，Jin Xinying，Zheng Yaohua，Li Yuexin，Chen Qianshan （ ，， ，）

Abstract：Inrecentyears，viralpneumoniahasrapidlyspreadgloballyAlthoughithasbeecontroledtosomeextent，thevirus's mutabilityandrapidtransmisionremainunpredictable.Newvariantsmayposefreshchallnges，necesitatingcontiuousmonitoring andassessmentepidemicrisks.Therefore，tosupportpublichealtheforts，thispaperproposesanimprovedCTimagedetection modelbasedonimprovedconvolutionalneuralntworksutilizingdeepleaingalgorithmstoasistphysiciansinrapidlydiagosing pneumonia.FirstthispapermodifiestheclasificationlayerAexNetandintroducesasymmetricconvolutionalblockstoeplace some itsconvolutionallayers.SecondatentionmodulesareembeddedatdiferentpositionswithinAlexNetandGoogLeNetto enhancemodelrecognitionperformance.Finally，transferlearningtechniquesareemployedfortrainingtheimprovedmodels.The overall accuracy AlexNet+SE+CBAM reaches 98.78% ，while GoogLeNet-CBAM achieves 99.22% .After ensemble learning on partial models，the ensemble model's overall accuracy improves to 99.54% with a recall rate 99.58% 三 Keywords：Deep Learning；Convolutional Neural Networks；Ensemble Learning；Pneumonia Diagnosis

0引言

2019年12月至2024年8月期间，世界卫生组织报告全球病毒性肺炎确诊病例共计776，007，137例，其中死亡病例7，059，612例。（剩余8385字）