Cross-Domain Vessel Segmentation via Latent Similarity Mining and Iterative Co-Optimization

2026-04-01

AI Summary

This work addresses the performance degradation in retinal vessel segmentation caused by domain shift between training and testing data by proposing a cross-domain segmentation method based on domain-invariant vascular prototypes. The approach constructs consistent vessel representations through deterministic latent space inversion and introduces a co-evolution mechanism that jointly optimizes a conditional diffusion generative model and a segmentation network. This joint optimization enables synergistic learning between image generation and segmentation tasks. Evaluated across multiple clinical datasets with substantial modality discrepancies, the method achieves state-of-the-art cross-domain segmentation performance, significantly enhancing the model’s generalization capability.

Abstract

Retinal vessel segmentation serves as a critical prerequisite for automated diagnosis of retinal pathologies. While recent advances in Convolutional Neural Networks (CNNs) have demonstrated promising performance in this task, significant performance degradation occurs when domain shifts exist between training and testing data. To address these limitations, we propose a novel domain transfer framework that leverages latent vascular similarity across domains and iterative co-optimization of generation and segmentation networks. Specifically, we first pre-train generation networks for source and target domains. Subsequently, the pretrained source-domain conditional diffusion model performs deterministic inversion to establish intermediate latent representations of vascular images, creating domain-agnostic prototypes for target synthesis. Finally, we develop an iterative refinement strategy where segmentation network and generative model undergo mutual optimization through cyclic parameter updating. This co-evolution process enables simultaneous enhancement of cross-domain image synthesis quality and segmentation accuracy. Experiments demonstrate that our framework achieves state-of-the-art performance in cross-domain retinal vessel segmentation, particularly in challenging clinical scenarios with significant modality discrepancies.

Problem

• cross-domain
• vessel segmentation
• domain shift
• retinal images
• domain generalization

Innovation

• cross-domain segmentation
• latent similarity mining
• iterative co-optimization
• conditional diffusion model
• domain adaptation