SSCA-Net: Simultaneous Self- and Channel-Attention Neural Network for Multiscale Structure-Preserving Vessel...

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Vessel segmentation is crucial for medical image analysis but challenging due to complex vessel structures.
• Existing methods struggle with tiny vessels and global spatial context.
• Accurate segmentation of multiscale vessels is essential for diagnosis.

Purpose of the Study:

• To propose a novel neural network, SSCA-Net, for multiscale structure-preserving vessel segmentation (MSVS).
• To improve the segmentation of intricate human vessel structures in medical imaging.
• To enhance the understanding of global semantic information in medical images.

Main Methods:

• Developed Simultaneous Self- and Channel-attention Neural Network (SSCA-Net).
• Integrated a Self- and Channel-Attention (SCA) module within the feature decoding stage.
• Employed self-attention for positional information and channel attention for global feature guidance.

Main Results:

• Achieved high performance on intracranial vessel segmentation (DSC: 96.21%, MIoU: 92.70%).
• Demonstrated superior preservation of vessel details and global spatial structures compared to state-of-the-art methods.
• Validated effectiveness across three benchmark vessel segmentation datasets.

Conclusions:

• SSCA-Net effectively addresses the MSVS problem by capturing multiscale structures and global context.
• The SCA mechanism significantly enhances the ability to segment complex vessel networks.
• SSCA-Net represents a significant advancement in medical image analysis for vessel segmentation.