Scholarlink

Lymph nodes, crucial components of the human immune system, filter lymph fluid and produce lymphocytes and antibodies to eliminate foreign cells and bacteria. However, diseases can alter their structure and morphology, often manifesting as superficial lymph node enlargement. Accurate assessment of lymph nodes is vital for medical diagnosis and treatment. Ultrasound, with its fast response, real-time monitoring, and high safety, is the mainstream imaging method for lymph node evaluation. However, ultrasound images often suffer from poor quality, making it challenging for doctors to distinguish between benign and malignant lymph nodes. In this study, we leverage deep learning and computer vision to classify and segment lymph node ultrasound images, aiding doctors in identifying lymph node areas and distinguishing between benign and malignant nodes. We collaborated with tertiary hospitals to obtain and annotate a dataset of 1,741 original ultrasound images. We tested the effectiveness of our optimized segmentation algorithm, R2GAB-Net, in lymph node segmentation tasks. Ablation experiments confirmed that adding a multi-head attention mechanism improves performance, and we optimized the Res2Att module with three attention heads for the network. Our results show that the introduction of the Res2Att module with the multihead attention mechanism significantly enhances the model’s segmentation ability. Compared to other advanced segmentation algorithms (U-Net, U-Net++, ViT, DeepLabv3+, ResUnet, FCN), R2GAB-Net demonstrated superior performance on lymph node ultrasound data. This innovative algorithm improves feature extraction and expression capabilities, leading to more accurate segmentation results. In conclusion, our study presents an optimized segmentation algorithm, R2GAB-Net, that effectively enhances lymph node ultrasound image segmentation. This technology has broad medical application prospects and significant social significance, aiding in the balanced development of medical resources and improving the utilization rate and coverage of medical services.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

R2GAB-Net for Deep Learning-Based Ultrasound Image Segmentation of Lymph Nodes

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jinghua Zhong,
  • Ying Fu,
  • Shi Tan,
  • Lifeng Guo,
  • Xueguan Yuan,
  • Yangan Zhang,
  • Xiaohong Huang

摘要

Lymph nodes, crucial components of the human immune system, filter lymph fluid and produce lymphocytes and antibodies to eliminate foreign cells and bacteria. However, diseases can alter their structure and morphology, often manifesting as superficial lymph node enlargement. Accurate assessment of lymph nodes is vital for medical diagnosis and treatment. Ultrasound, with its fast response, real-time monitoring, and high safety, is the mainstream imaging method for lymph node evaluation. However, ultrasound images often suffer from poor quality, making it challenging for doctors to distinguish between benign and malignant lymph nodes. In this study, we leverage deep learning and computer vision to classify and segment lymph node ultrasound images, aiding doctors in identifying lymph node areas and distinguishing between benign and malignant nodes. We collaborated with tertiary hospitals to obtain and annotate a dataset of 1,741 original ultrasound images. We tested the effectiveness of our optimized segmentation algorithm, R2GAB-Net, in lymph node segmentation tasks. Ablation experiments confirmed that adding a multi-head attention mechanism improves performance, and we optimized the Res2Att module with three attention heads for the network. Our results show that the introduction of the Res2Att module with the multihead attention mechanism significantly enhances the model’s segmentation ability. Compared to other advanced segmentation algorithms (U-Net, U-Net++, ViT, DeepLabv3+, ResUnet, FCN), R2GAB-Net demonstrated superior performance on lymph node ultrasound data. This innovative algorithm improves feature extraction and expression capabilities, leading to more accurate segmentation results. In conclusion, our study presents an optimized segmentation algorithm, R2GAB-Net, that effectively enhances lymph node ultrasound image segmentation. This technology has broad medical application prospects and significant social significance, aiding in the balanced development of medical resources and improving the utilization rate and coverage of medical services.