Multi-Layer Preprocessing and U-Net with Residual Attention Block for Retinal Blood Vessel Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Ophthalmology

Background:

• Retinal blood vessel segmentation is crucial for diagnosing eye conditions like glaucoma and macular degeneration.
• Existing methods face challenges with image noise and data limitations.

Purpose of the Study:

• To develop and evaluate a robust framework for accurate retinal blood vessel segmentation.
• To improve diagnostic support for various ocular diseases through enhanced image analysis.

Main Methods:

• A two-stage framework involving multi-layer preprocessing and U-Net segmentation with attention.
• Preprocessing includes noise reduction (CNN with MF, D_U-Net), data imputation, and data augmentation (LDM).
• Segmentation utilizes a U-Net with a multi-residual attention block for precise vessel identification.

Main Results:

• The framework achieved high performance metrics: Dice score (95.32%), accuracy (93.56%), precision (95.68%), and recall (95.45%).
• Effective noise reduction was demonstrated via PSNR and SSIM values across various noise levels.
• Latent Diffusion Model (LDM) showed strong performance in data augmentation with an inception score of 13.6 and FID of 46.2.

Conclusions:

• The proposed framework significantly enhances retinal blood vessel segmentation accuracy.
• The multi-stage approach effectively addresses preprocessing challenges, leading to reliable diagnostic information.
• This method offers a promising tool for clinical ophthalmology and research.