Mamba-based context-aware local feature network for vessel detail enhancement.

Area of Science:

• Medical imaging
• Biomedical engineering
• Computer vision

Background:

• Accurate blood vessel analysis is crucial for clinical diagnosis and intervention.
• Near-infrared-II (NIR-II) fluorescence imaging offers advanced visualization but struggles with low-contrast deep vessels.
• Vessel segmentation in low-contrast regions remains a significant challenge in medical imaging.

Purpose of the Study:

• To develop an advanced network for precise blood vessel segmentation, particularly in challenging low-contrast environments.
• To enhance the utility of NIR-II fluorescence imaging for detailed vascular analysis.
• To improve the accuracy of identifying fine vascular structures for clinical applications.

Main Methods:

• Proposed CALFNet, a UNet-like architecture incorporating a ResNet encoder and a Mamba-based context-aware module.
• Utilized global vessel contextual information to improve segmentation in low-contrast areas.
• Implemented a feature-enhance module to preserve and refine local vascular details.

Main Results:

• CALFNet demonstrated superior performance compared to existing methods on both NIR-II and visible light retinal datasets.
• Achieved more accurate vessel segmentation, especially in regions with low contrast.
• Showcased enhanced robustness in segmenting fine vascular structures.

Conclusions:

• CALFNet effectively segments vessels in low-contrast regions, outperforming comparison methods.
• The network enhances the capabilities of NIR-II fluorescence imaging for vascular analysis.
• CALFNet provides valuable support for clinical diagnosis and medical intervention through improved imaging accuracy.