基于改进TransU-Net的乳腺肿瘤分割算法研究
Research on breast tumor segmentation algorithm based on improved TransU-Net
朱盛滔 1贺泽民 1陈超峰1
作者信息
摘要
针对超声乳腺肿瘤图像中存在的高散斑噪声较多、肿瘤边缘模糊以及形状复杂多样等问题,文章在TransU-Net的基础上进行改进,提出了基于TransU-Net的多路径特征融合网络(MFF-Net).文章分析了MFF-Net的整体结构、多路径特征融合提取模块以及深监督机制,通过实验验证了MSF-Net在处理边缘模糊和形状复杂多样的乳腺超声图像方面的有效性.结果显示,MSF-Net在多个评价指标上优于现有的主流方法.
Abstract
Aiming at the problems of high scattering noise,blurred tumor edges and complex and diverse shapes in ultrasound breast tumor images,this paper proposes a TransU-Net-based multipath feature fusion network(MFF-Net)by improving on the basis of TransU-Net.In this paper,we analyze the overall structure of MFF-Net,the multipath feature fusion extraction module and the deep supervision mechanism,and experimentally verified the effectiveness of MSF-Net in processing breast ultrasound images with blurred edges and complex and diverse shapes.The results show that MSF-Net outperforms existing mainstream methods in several evaluation indexes.
关键词
乳腺超声图像分割/深度学习/多路径融合/深监督Key words
breast ultrasound image segmentation/deep learning/multipath fusion/deep supervision引用本文复制引用
出版年
2024
NETL
NSTL
万方数据