Multi-Level Attention-Guided U-Net with Hierarchical Fusion for Retinal Vessel Extraction

Description:

Retinal blood vessels form a complex vascular network that can provide important information about the structural characteristics of the eye. Accurate identification of these vessels from retinal fundus images is an important task in ophthalmic image analysis and computer-aided diagnosis. However, variations in vessel thickness, illumination, contrast, image quality, and the presence of fine or low-contrast vessels make automatic vessel extraction challenging. Retinal vessel segmentation aims to separate vascular structures from the surrounding retinal background at the pixel level, providing a detailed representation of the retinal vascular network. Such segmentation can support subsequent analysis of vessel morphology, including vessel width, branching patterns, tortuosity, and vascular distribution, and can serve as an important preprocessing step for automated retinal disease analysis

Input: Retinal fundus images

Output: Pixel-level retinal vessel segmentation masks

Tags:

#RetinalImaging, #RetinalVesselSegmentation, #MedicalImageSegmentation, #FundusImages, #omputerVision, #eepLearning, #BloodVesselDetection, #ImageProcessing, #Healthcare

References:

1. Johny, D., Chandan, V., Loni, S., Devika Gireesh, A., Saimahesh, V., George, K., ... & Jayaraj, P. B. (2026). Structure-aware learnable multi-scale attention for retinal vessel analysis. Scientific Reports.

2. Ling, Z., Yu, J., Zuo, Q., & Lei, B. (2026). Multi-stage cascaded refinement with wavelet downsampling for retinal vessel segmentation. Biomedical Signal Processing and Control, 112, 108824.