安徽大学学报（自然科学版）2025，Vol.49Issue(5)：19-28,10.DOI:10.3969/j.issn.1000-2162.2025.05.003
基于增强跳跃连接U-Net的肾脏肿瘤分割方法
Segmentation of kidney tumors based on enhanced skip-connection U-Net
摘要
Abstract
Computed tomography(CT)occupies an important position in the diagnosis of kidney tumors,and accurate detecting tumor detection in CT images can effectively assist doctors in the preliminary diagnosis.Aiming at the problems of diverse scales and ambiguous edges of tumors in CT images,this paper proposed a kidney tumor segmentation method based on enhanced skip-connection U-Net.Based on nnU-Net(no-new-net),this method designed a parallel residual feature enhancer in the skip-connection part to realize the effective extraction and enhancement of various scale features and edge features of kidney tumors.Specifically,a multi-branch parallel convolution structure was designed in the parallel residual feature enhancer to obtain the rich features of the tumor,while the parallel convolution was set up with residual connections to extract the multi-scale features of the tumor,and the depth-separable convolution was improved to enhance the edge feature information.Experimental results on the constructed CT image dataset of kidney tumors showed the superiority of the method proposed in this paper in achieving the DSC(dice similarity coefficient)of 86.36％and 86.96％for kidney tumors,which were 1.31％and 0.96％higher than the baseline,respectively,by using the two training strategies of five-fold cross-validation and all-data training.关键词
CT图像/深度学习/肾脏肿瘤分割/增强跳跃连接/并行残差特征增强器Key words
CT images/deep learning/kidney tumor segmentation/enhanced skip-connection/parallel residual feature enhance分类
信息技术与安全科学引用本文复制引用
鲍文霞,杜银徕,姚文君,朱宏庆..基于增强跳跃连接U-Net的肾脏肿瘤分割方法[J].安徽大学学报（自然科学版）,2025,49(5):19-28,10.基金项目
安徽省转化医学研究院科研基金(2021zhyx-C45) （2021zhyx-C45）
