Do Not Break the Vessels: Structure-Preserving Mean Flow for Vascular Image Translation

2026-06-29

AI Summary

Existing cross-modal vascular image translation methods often compromise anatomical continuity and introduce artifacts due to insufficient preservation of topological structure, thereby undermining clinical reliability. To address this, this work proposes a topology-preserving transport framework that decouples appearance variation from structural deformation through orthogonal vector field constraints. The method explicitly enforces vascular topological consistency at every step of a Brownian bridge diffusion process and incorporates a Prototype-Guided Structure Refinement (PGSR) module to enhance inference quality. Notably, it is the first diffusion-based approach to guarantee structural invariance during translation. Experimental results demonstrate superior performance, achieving PSNR values of 24.96 dB and 24.83 dB on the NIRII-to-2PF and fundus datasets, respectively—significantly outperforming existing methods while effectively preserving vascular continuity and anatomical fidelity.

Abstract

Reconstructing anatomically faithful vascular structures from clinically accessible imaging modalities is of substantial clinical significance. However, existing cross-modal translation methods mainly emphasize pixel-level fidelity or visual realism and treat structure preservation as a property of the final output rather than an invariant of the generative process. This limitation often leads to structural discontinuities and artifacts, compromising anatomical coherence and clinical reliability. In this work, we propose a Structure-Preserving Mean Flow (SPMF) framework that formulates vascular image translation as a topology-invariant transport process. Based on a structural invariance principle, we derive an orthogonality constraint on the flow velocity field that formally separates appearance transport from topological distortion. We implement this constraint as a time-weighted surrogate objective within a Brownian bridge diffusion model to preserve topology at every diffusion step. Moreover, we propose a Prototype-Guided Structural Refinement (PGSR) module to align degraded inference-time structures with reliable training-time structures. Experiments on paired NIRII-to-2PF and fundus datasets demonstrate consistent improvements over state-of-the-art methods, achieving peak PSNR values of 24.96 dB and 24.83 dB, respectively.

Problem

• vascular image translation
• structure preservation
• topological distortion
• anatomical coherence
• cross-modal translation

Innovation

• Structure-Preserving Mean Flow
• topology-invariant transport
• orthogonality constraint
• Brownian bridge diffusion
• Prototype-Guided Structural Refinement