重庆理工大学学报2025，Vol.39Issue(9)：107-113,7.DOI:10.3969/j.issn.1674-8425(z).2025.05.013
深度特征融合驱动的肺腺癌分类算法研究
Research on lung adenocarcinoma classification algorithms driven by deep feature fusion
摘要
Abstract
A deep model-based adaptive fusion algorithm,ViT-CLSNet,is proposed for lung adenocarcinoma subtype classification.The method integrates ViT and DenseNet in a parallel manner to construct a dual-stream network,enhancing the modeling capability for heterogeneous features.Additionally,a novel High-Order Coordinate Attention Mechanism(HOCAM)is designed to effectively localize local glandular structures in lung adenocarcinoma.Furthermore,an Adaptive Fusion Block(AFB)is introduced to improve the adaptability of feature fusion in the dual-stream network.Experimental results demonstrate ViT-CLSNet achieves an average classification accuracy of 90.92%on a lung adenocarcinoma dataset,outperforming the other best-performing network by 3.10%.关键词
病理图像/图像分类/模型融合/注意力机制Key words
pathological images/image classification/model fusion/attention mechanism分类
临床医学引用本文复制引用
刘亚楠,孟赋涵,罗家洋,冯鹏..深度特征融合驱动的肺腺癌分类算法研究[J].重庆理工大学学报,2025,39(9):107-113,7.基金项目
重庆开放大学(重庆工商职业学院)科研项目(NDYB2023-07) （重庆工商职业学院）
重庆市科委技术创新与应用发展专项(cstc2021 jscx-gksbX0056) （cstc2021 jscx-gksbX0056）
中央高校基本科研业务费项目(2023CDJKYJH085) （2023CDJKYJH085）
