Fu Ying, Tan Shi, Kadoch Michel, Zhong Jinghua, Guo Lifeng, Zhang Yangan, Huang Xiaohong, Yuan Xueguang
Department of Ultrasound, Peking University Third Hospital, Beijing 100191, China.
Department of Electrical Engineering, École de Technologie Supérieure, Université du Québec, Québec, QC J5A 0M3, Canada.
Bioengineering (Basel). 2025 Feb 16;12(2):190. doi: 10.3390/bioengineering12020190.
This study presents a novel Semantic-Attention Enhanced Dynamic Swin Convolutional Block Attention Module(CBAM) Transformer (DSC-Transformer) for lymph node ultrasound image classification. The model integrates semantic feature extraction and multi-scale attention mechanisms with the Swin Transformer architecture, enabling efficient processing of diagnostically significant regions while suppressing noise. Key innovations include semantic-driven preprocessing for localized diagnostic focus, adaptive compression for bandwidth-limited scenarios, and multi-scale attention modules for capturing both global anatomical context and local texture details. The model's effectiveness is validated through comprehensive experiments on diverse datasets and Grad-Channel Attention Module (CAM) visualizations, demonstrating superior classification performance while maintaining high efficiency in remote diagnostic settings. This semantic-attention enhancement makes the DSC-Transformer particularly effective for telemedicine applications, representing a significant advancement in AI-driven medical image analysis with broad implications for telehealth deployment.
本研究提出了一种用于淋巴结超声图像分类的新型语义注意力增强动态Swin卷积块注意力模块（CBAM）Transformer（DSC-Transformer）。该模型将语义特征提取和多尺度注意力机制与Swin Transformer架构相结合，能够在抑制噪声的同时高效处理具有诊断意义的区域。关键创新包括用于局部诊断聚焦的语义驱动预处理、用于带宽受限场景的自适应压缩以及用于捕获全局解剖背景和局部纹理细节的多尺度注意力模块。通过在不同数据集上进行的综合实验和Grad-通道注意力模块（CAM）可视化验证了该模型的有效性，证明了其在保持远程诊断设置高效率的同时具有卓越的分类性能。这种语义注意力增强使得DSC-Transformer在远程医疗应用中特别有效，代表了人工智能驱动的医学图像分析的重大进展，对远程医疗部署具有广泛影响。
