RelativeFlow: Taming Medical Image Denoising Learning with Noisy Reference

2026-04-16

AI Summary

Medical image denoising is fundamentally constrained by the absence of absolutely clean reference images, a limitation known as the noise-referenced problem, which hinders optimal performance. To address this challenge, this work proposes RelativeFlow, a novel framework that reformulates absolute denoising as a relative mapping between noisy images. By leveraging Consistent Transport (CoT) and a simulation-based Velocity Field (SVF), RelativeFlow guides inputs of arbitrary quality toward a unified high-quality target without requiring clean-image supervision. The approach enables unified learning under heterogeneous noise conditions and is compatible with diverse medical imaging modalities. Extensive experiments demonstrate that RelativeFlow significantly outperforms existing methods in both CT and MRI denoising tasks, effectively resolving the long-standing difficulty of noise-referenced medical image restoration.

Technology Category

• Computer Vision: Medical and Biological Imaging
• Machine Learning: Multimodal Learning
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

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.

Problem

• Medical image denoising
• Noisy reference
• Supervision
• Noise assumptions
• Denoising performance

Innovation

• RelativeFlow
• flow matching
• noisy reference
• medical image denoising
• simulation-based velocity field