Placental Super Micro-vessels Segmentation Based on ResNeXt with Convolutional Block Attention and U-Net

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Pathology

Background:

• Accurate segmentation of placenta super micro-vessels is crucial for diagnosing placental diseases.
• Current automatic segmentation algorithms suffer from information redundancy and low utilization, limiting segmentation accuracy.

Purpose of the Study:

• To propose a novel RC-UNet model for improved placenta super micro-vessel segmentation.
• To address the limitations of existing algorithms in terms of information redundancy and utilization.

Main Methods:

• Developed a ResNeXt with convolutional block attention module (CBAM) and UNet (RC-UNet) model.
• Utilized UNet as the backbone for initial feature extraction.
• Employed ResNeXt-CBAM as an attention module for feature refinement and weighting, employing a split-transform-merge strategy.

Main Results:

• The proposed RC-UNet model demonstrated superior segmentation results compared to other algorithms.
• Achieved enhanced segmentation accuracy for anatomical structures including umbilical cord blood (UC), stem villus (ST), and maternal blood (MA).
• The model effectively reduced hyperparameter redundancy and improved information utilization through CBAM processing.

Conclusions:

• The RC-UNet model offers a significant advancement in placenta super micro-vessel segmentation.
• This improved segmentation accuracy aids in the diagnosis of placental diseases.
• The method effectively enhances information utilization, leading to better diagnostic outcomes.