Landmark-Assisted Anatomy-Sensitive Retinal Vessel Segmentation Network

Area of Science:

• Medical Imaging
• Computer Vision
• Ophthalmology

Background:

• Automatic retinal vessel segmentation is crucial for diagnosing eye diseases.
• Current deep learning methods struggle with thin vessel segmentation and maintaining vessel connectivity.

Purpose of the Study:

• To develop a novel anatomy-sensitive framework for enhanced retinal vessel segmentation.
• To improve the detection of thin vessels and preserve the topological continuity of the vasculature.

Main Methods:

• Utilized TransUNet as the backbone architecture.
• Incorporated self-supervised extracted anatomical landmarks guided by contrastive learning.
• Employed a framework sensitive to anatomical structures for network guidance.

Main Results:

• Achieved superior performance on DRIVE and CHASE-DB1 datasets, outperforming state-of-the-art methods.
• Demonstrated competitive results on the STARE dataset.
• Visualizations confirmed improved topological continuity and thin vessel identification.

Conclusions:

• The proposed anatomy-sensitive framework effectively addresses limitations in current retinal vessel segmentation.
• The method shows significant potential for clinical applications in ophthalmic disease diagnosis.
• Self-supervised landmark guidance enhances morphological feature learning for improved segmentation accuracy.