RU-Net: An improved U-Net placenta segmentation network based on ResNet. (December 2022)

• Record Type: Journal Article
• Authors: Wang, Yi; Li, Yuan-Zhe; Lai, Qing-Quan; Li, Shu-Ting; Huang, Jing
• Is Part Of: Computer methods and programs in biomedicine. Volume 227(2022)
• Publication Date: 2022-12
• Subjects: U-Net -- ResNet -- Placenta segmentation -- Deep learning -- Semantic segmentation; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2022.107206
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: An improved U-Net framework named RU-Net is proposed. Direct mapping structure of ResNet is added to the original contraction path and expansion path of U-Net. Residual structure recovers the feature information of the image to improve segmentation accuracy. The proposed model has better segmentation performance when compared with other methods. Our segmentation model can be used for accurate placenta in practical clinical application. Abstract: Background: In recent years, with the increase of late puerperium, cesarean section and induced abortion, the incidence of placenta accreta has been on the rise. It has become one of the common clinical diseases in obstetrics and gynecology. In clinical practice, accurate segmentation of placental tissue is the basis for identifying placental accreta and assessing the degree of accreta. By analyzing the placenta and its surrounding tissues and organs, it is expected to realize automatic computer segmentation of placental adhesion, implantation, and penetration and help clinicians in prenatal planning and preparation. Methodology: We propose an improved U-Net framework: RU-Net. The direct mapping structure of ResNet was added to the original contraction path and expansion path of U-Net. The feature information of the image was restored to a greater extent through the residual structure to improve the segmentation accuracy of the image. Results: Through testing on the collected placenta dataset, it is found that our proposed RU-Net network achieves 0.9547 and 1.32% on the Dice coefficient and RVD index, respectively. We also compared with the segmentation frameworks of other papers, and the comparison results show that our RU-Net network has better performance and can accurately segment the placenta. Conclusion: Our proposed RU-Net network addresses issues such as network degradation of the original U-Net network. Good segmentation results have been achieved on the placenta dataset, which will be of great significance for pregnant women's prenatal planning and preparation in the future.