Improved nested U-structure for accurate nailfold capillary segmentation.

Area of Science:

• Biomedical Imaging
• Medical Diagnostics
• Computational Pathology

Background:

• Nailfold capillaries provide insights into various diseases.
• Accurate segmentation of nailfold capillaries is crucial for diagnosis.
• Current methods struggle with precise capillary-background separation.

Purpose of the Study:

• To develop an accurate nailfold capillary image segmentation method.
• To improve the extraction of morphological information from nailfold capillaries.
• To enhance clinical diagnosis and research for capillary-related diseases.

Main Methods:

• Proposed a novel segmentation method integrating U²-Net and Transformer.
• Developed a decoder-encoder network with Transformer layers in a U²-Net architecture.
• Utilized the network to extract multiscale and multilevel features for high-resolution maps.

Main Results:

• Achieved 98.23% overall accuracy, 88.56% Dice coefficient, and 80.41% IoU.
• Demonstrated significant improvements over U²-Net, Res-Unet, and U-Net.
• The Transformer-U²Net method provided detailed and accurate segmented capillary structures.

Conclusions:

• The Transformer-U²Net network excels in nailfold capillary segmentation.
• Accurate segmentation aids in the precise diagnosis and treatment of related diseases.
• This method offers a valuable tool for clinical practice and research.