RU-Net: An improved U-Net placenta segmentation network based on ResNet.

Area of Science:

• Medical Imaging Analysis
• Deep Learning in Obstetrics
• Computational Pathology

Background:

• The incidence of placenta accreta is increasing due to factors like cesarean sections and induced abortions.
• Accurate placental tissue segmentation is crucial for diagnosing placenta accreta and assessing its severity.
• Computer-aided segmentation can assist clinicians in prenatal planning and preparation.

Purpose of the Study:

• To develop an improved deep learning framework for accurate placental tissue segmentation.
• To enhance the segmentation accuracy of the U-Net architecture for clinical applications.

Main Methods:

• An improved U-Net framework, termed RU-Net, was proposed.
• The RU-Net incorporates the direct mapping structure of ResNet into the U-Net's contraction and expansion paths.
• Residual structures were utilized to restore image feature information and improve segmentation accuracy.

Main Results:

• The RU-Net achieved a Dice coefficient of 0.9547 and a Relative Volume Difference (RVD) index of 1.32% on a placenta dataset.
• Comparative analysis demonstrated superior performance of RU-Net over other segmentation frameworks.
• The network accurately segmented placental tissue.

Conclusions:

• The RU-Net effectively addresses limitations such as network degradation found in the original U-Net.
• The proposed network yields excellent segmentation results for placental tissue.
• This technology holds significant potential for improving prenatal planning and preparation for pregnant women.