RelativeFlow: Taming Medical Image Denoising Learning with Noisy Reference

TL;DR

RelativeFlow introduces a flow matching framework that effectively learns from noisy references in medical image denoising, outperforming existing methods across CT and MR modalities.

Contribution

It reformulates flow matching by decomposing the noise-to-clean mapping into relative mappings and introduces components to handle heterogeneous noisy references.

Findings

Significantly outperforms existing MID methods on CT and MR datasets.

Effectively learns from heterogeneous noisy references without clean supervision.

Demonstrates robustness across different medical imaging modalities.

Abstract

Medical image denoising (MID) lacks absolutely clean images for supervision, leading to a noisy reference problem that fundamentally limits denoising performance. Existing simulated-supervised discriminative learning (SimSDL) and simulated-supervised generative learning (SimSGL) treat noisy references as clean targets, causing suboptimal convergence or reference-biased learning, while self-supervised learning (SSL) imposes restrictive noise assumptions that are seldom satisfied in realistic MID scenarios. We propose \textbf{RelativeFlow}, a flow matching framework that learns from heterogeneous noisy references and drives inputs from arbitrary quality levels toward a unified high-quality target. RelativeFlow reformulates flow matching by decomposing the absolute noise-to-clean mapping into relative noisier-to-noisy mappings, and realizes this formulation through two key components: 1) consistent transport (CoT), a displacement map that constrains relative flows to be components of and progressively compose a unified absolute flow, and 2) simulation-based velocity field (SVF), which constructs a learnable velocity field using modality-specific degradation operators to support different medical imaging modalities. Extensive experiments on Computed Tomography (CT) and Magnetic Resonance (MR) denoising demonstrate that RelativeFlow significantly outperforms existing methods, taming MID with noisy references.