NGPS: Structure-Preserving Self-Supervised Denoising via Neighbor-Guided Patch Sampling

2026-06-22

AI Summary

This work addresses the challenge of anatomical inconsistency in medical images caused by misalignment between adjacent slices. To this end, the authors propose a lightweight Neighborhood-Guided Patch Sampling (NGPS) framework for structure-preserving self-supervised denoising. The method innovatively decouples structural alignment from supervision signal retrieval without requiring explicit image registration. Specifically, it employs bilateral filtering to generate guidance images for local structural alignment and matches image patches within a neighborhood based on structural similarity. High-frequency details are then directly extracted from the original noisy slices as supervision signals, thereby circumventing information loss typically induced by masking strategies. Experimental results demonstrate that NGPS significantly improves image fidelity, edge sharpness, and anatomical consistency on both clinical CT and synthetic Rician-noise MRI data.

Abstract

Neighboring-slice self-supervised denoising is attractive for volumetric medical imaging, yet inter-slice misalignment breaks anatomical correspondence and often yields ghosting and blurred margins when adjacent slices are used naively as targets. We propose Neighbor-Guided Patch Sampling (NGPS), a lightweight framework that constructs neighboring supervision under local inter-slice misalignment without explicit registration. To avoid learning from misleading targets, prior methods commonly mask discrepant regions, but this stabilizes training at the cost of leaving a non-trivial portion of neighboring evidence unexploited, particularly around high-frequency anatomical boundaries. NGPS addresses this by decoupling structure matching from signal retrieval: for each masked location, it searches a local neighborhood for structurally similar candidate patches using a simple guide image (e.g., fast bilateral filtering), while retrieving the supervision signal directly from the raw noisy neighbor at the matched coordinates. By matching on a noise-attenuated guide while retrieving raw values from neighboring slices, NGPS constructs local pseudo targets without a learned registration module. Across the evaluated CT and synthetic-Rician MRI settings, NGPS improves fidelity and structure-sensitive metrics. Code is available at https://github.com/cv-cho/NGPS .

Problem

• self-supervised denoising
• inter-slice misalignment
• volumetric medical imaging
• anatomical correspondence
• ghosting artifacts

Innovation

• self-supervised denoising
• neighbor-guided patch sampling
• structure-preserving
• inter-slice misalignment
• medical image denoising