ESDiff: a joint model for low-quality retinal image enhancement and vessel segmentation using a diffusion model.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate blood vessel extraction from fundus images is crucial for clinical disease diagnosis.
• Clinical fundus images frequently exhibit quality degradation due to uneven illumination, blur, and artifacts.

Purpose of the Study:

• To propose a unified framework, ESDiff, for integrated retinal image enhancement and vessel segmentation.
• To address challenges posed by low-quality fundus images in clinical screening.

Main Methods:

• Developed a novel diffusion model-based framework (ESDiff) for image enhancement.
• Incorporated mask refinement as an auxiliary task using a vessel mask-aware diffusion model.
• Utilized a modified UNet with low-quality images and illumination maps to derive degradation factors for enhancing intermediate diffusion model results.

Main Results:

• ESDiff effectively integrates retinal image enhancement and vessel segmentation.
• The framework preserves pathological features and pertinent information despite image degradations.
• Experiments on DRIVE, STARE, CHASE_DB1, and EyeQ datasets show ESDiff outperforms state-of-the-art methods.

Conclusions:

• ESDiff offers a robust solution for improving the quality of retinal fundus images.
• The proposed method enhances the accuracy of blood vessel segmentation in challenging clinical scenarios.
• ESDiff demonstrates significant potential for advancing automated disease diagnosis through improved image analysis.