Improving Retinal Artery-Vein Segmentation via Geometric Energy Fields

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal artery/vein (A/V) segmentation is crucial for diagnosing eye diseases.
• Existing methods often struggle with robustness and structural consistency.
• Local appearance cues are insufficient for precise vascular segmentation.

Purpose of the Study:

• To develop a novel Geometric Energy Field (GEF) supervision framework.
• To enhance the robustness and structural consistency of retinal A/V segmentation.
• To improve the clinical plausibility of segmentation results.

Main Methods:

• Introduction of two geometrically complementary energy fields: Distance Energy Field (DEF) and Orientation Energy Field (OEF).
• DEF models pixel-wise proximity to arteries and veins, capturing spatial coupling.
• OEF enforces vessel elongation and directional continuity, suppressing label flipping.

Main Results:

• The GEF framework achieved superior A/V segmentation accuracies across six datasets (ranging from 97.1% to 99.1%).
• Outperformed state-of-the-art methods in retinal A/V segmentation tasks.
• Demonstrated enhanced coherence, stability, and clinical plausibility of segmentation results.

Conclusions:

• The proposed GEF supervision framework significantly improves retinal A/V segmentation.
• Geometry-aware energy fields provide effective guidance beyond local appearance.
• The method offers a more robust and reliable tool for clinical applications.