| 摘要: |
| 为了正确诊断肺癌转移，本文应用深度学习技术对肺癌患者颈部淋巴结超声图像病灶区域分割，提出了一种用于超声图像分割的级联注意力UNet网络，该级联结构是将注意力UNet与EfficientNet相结合的二阶段分割网络，第一阶段为粗分割，第二阶段为细分割，编码器采用EfficientNet-B5作为主干网，图像多尺度输入；提出了适用于小目标、小样本场景的新损失函数；实验结果表明，该文提出的级联结构网络在颈部淋巴结超声图像分割中网络性能优异，Dice系数达到0.95，较其他UNet方法具有更优的分割性能。 |
| 关键词: 图像分割 超声图像 注意力机制 级联 |
| DOI： |
| 投稿时间：2022-03-03修订日期：2022-04-20 |
| 基金项目:上海市卫生健康委员会科研课题面上项目 |
|
| RESEARCH ON ULTRASONIC IMAGE SEGMENTATION OF CERVICAL LYMPH NODES IN LUNG CANCER PATIENTS BASED ON CASCADE STRUCTURE |
| gongxia,wu weihua |
| () |
| Abstract: |
| In order to correctly diagnose the metastasis of lung cancer, this paper applies deep learning technology to segment the focus area of cervical lymph node ultrasound image of lung cancer patients , and proposes a cascade attention UNet network for ultrasound image segmentation. The cascade structure is a two-stage segmentation network combining attention UNet and EfficientNet. The segmentation model includes one-stage coarse segmentation and two-stage fine segmentation. The encoder uses EfficientNet-B5 as the backbone network. The multi-scale features of the image are taken as the input. A new loss function is proposed, which is suitable for small target and few-shot scenarios. The experimental results show that the proposed cascade structure has excellent network performance in cervical lymph node ultrasonic image segmentation, and the Dice coefficient reaches 0.95, which has better segmentation performance than other UNet methods. |
| Key words: |
