A processing method of medical image segmentation based on U-Net network with residual connection under single source domain environment
10.3969/j.issn.1672-8270.2025.05.005
- VernacularTitle:单源域环境下基于残差连接U-Net网络的医学影像分割处理方法研究
- Author:
Yushan ZENG
1
;
Yan XU
1
;
Siyu MA
1
;
Haiqing XU
1
;
Chundong QIU
1
Author Information
1. 南京鼓楼医院临床医学工程处 南京 210000
- Publication Type:Journal Article
- Keywords:
Single source domain environment;
Residual connection;
Medical image;
Data amplification;
Image features;
U-net network
- From:
China Medical Equipment
2025;22(5):22-27,32
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To propose a processing method of medical image segmentation based on U-Net network with residual connection under single source domain environment,so as to improve the limited diversity of data samples of medical image in the single-source domain environment.Methods:The medical image data under the single-source domain environment were amplified through image transformation,contrast adjustment,and noise addition.The U-Net network was used as the basic structure,and multiple residual connection modules were introduced to achieve reuse of feature.And then,the features at shallow layer were directly transmitted to deep layer,and features at different levels were utilized for segmentation.The input features were added to the features that were processed by partial network layers,thus better learned the features in the data,and constructed medical image segmentation model based on the U-Net network with residual connection.The segmentation model was trained by using the gradient descent method,and the combination of that and the Generalized Dice Loss(GDL)function.Results:The processing method of medical image segmentation based on the U-Net network can effectively achieve segmentation of medical image.Even under low signal-to-noise ratio conditions and facing to different types of segmentation tasks of medical image,the DSC value was greater than 0.90.Conclusion:The processing method of medical image segmentation based on the U-Net network in this study can effectively enhance recognition ability for the features of complex lesions,and increase the segmentation precise of medical images,and improve the accuracy of clinical diagnosis and the effectiveness of treatment.
