ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image

2026-07-27

AI Summary

This work addresses the critical challenge of extreme annotation scarcity in retinal vessel segmentation by achieving high-performance segmentation under an unprecedented semi-supervised setting that uses only a single annotated image. The proposed method leverages DINOv3 features to transfer vascular cues from the source to the target domain, constructs multi-granularity vessel prototypes, and integrates physical priors to generate initial pseudo-labels. These pseudo-labels are subsequently refined through a weighted training scheme and adversarial optimization, all embedded within the Mask2Former framework. Evaluated on eight public datasets, the approach achieves state-of-the-art Dice and clDice scores on six datasets and consistently obtains the lowest HD95 across all datasets. Remarkably, it retains 93.7% of Dice and 95.1% of clDice performance relative to fully supervised models, substantially reducing reliance on large-scale annotated data.

Abstract

Learning from minimal human supervision is a long-standing goal in medical image analysis, where dense expert annotations are costly. We study retinal vessel segmentation in an extreme semi-supervised setting with one annotated image and a pool of unlabeled images. We propose ESRVS, which selects a representative reference image for manual annotation and transfers vessel cues using target-domain-adapted DINOv3 features. ESRVS constructs a multi granular vessel prototype, combines prototype-similarity maps with a physics-inspired prior to generate initial pseudo-labels, and refines the transferred supervision through weighted pseudo-label training and adversarial refinement. Across eight public datasets, ESRVS achieves the best Dice and clDice on six datasets, and the best HD95 on all eight datasets among the compared semi-supervised methods, although those methods use 10 to 20% labeled data. With Mask2Former, ESRVS retains on average 93.7% of fully supervised Dice and 95.1% of fully supervised clDice. These results demonstrate the potential of foundation-model label propagation for highly label-efficient retinal vessel segmentation. Code is available at https://github.com/IAANNH/ESRVS.

Problem

• retinal vessel segmentation
• extreme semi-supervised learning
• medical image analysis
• label efficiency
• pseudo-labeling

Innovation

• extreme semi-supervised learning
• retinal vessel segmentation
• foundation model
• pseudo-label refinement
• label-efficient learning