机器人外科学杂志（中英文）2025，Vol.6Issue(9)：1454-1460,7.DOI:10.12180/j.issn.2096-7721.2025.09.003
融合HRNet模块和可变形注意力机制的乳腺癌图像分类技术研究
Research on breast cancer image classification technology integrating HRNet module and deformable attention mechanism
摘要
Abstract
Objective:To enhance the accuracy of breast cancer image classification and diagnosis by integrating high-resolution network(HRNet)modules and dynamic deformable attention mechanisms into the U-Net model.Methods:Based on U-Net,the proposed model incorporates the powerful HRNet module with atrous spatial pyramid pooling(ASPP)to capture multi-scale features.It also integrates global multi-scale attention mechanisms,multi-scale residual convolution modules,and dynamic deformable attention mechanisms to improve the model's focus on critical features.Results:The improved U-Net-HRNet model outperforms existing breast cancer image classification methods in key metrics including accuracy,recall,precision,F1-measure,Jaccard index,and Dice similarity coefficient.Ablation experiments indicate significant performance degradation when any module is removed,with the HRNet module having the most substantial impact.When the HRNet module was removed,accuracy,F1-measure,Jaccard index,and Dice similarity coefficient decreased by 6.9%,7.8%,11.7%,and 14.1%,respectively.Conclusion:The enhanced U-Net-HRNet model has superior performance in breast cancer image classification and provides effective support for clinical diagnosis.关键词
乳腺癌/图像分类/HRNet/空洞空间金字塔池化/动态可变形注意力Key words
Breast Cancer/Image Classification/HRNet/Atrous Spatial Pyramid Pooling/Dynamic Deformable Attention分类
医药卫生引用本文复制引用
田兄玲,宋丽俊,郭辉,王贝..融合HRNet模块和可变形注意力机制的乳腺癌图像分类技术研究[J].机器人外科学杂志（中英文）,2025,6(9):1454-1460,7.基金项目
新疆人工智能影像辅助诊断重点实验室项目(XJRG2N2024009)Xinjiang Key Laboratory of Artificial Intelligence for Medical Imaging Assisted Diagnosis Project(XJRG2N2024009) （XJRG2N2024009）
