---
title: "UW-Madison's SGPnP Beats Diffusion Solvers on Severely Corrupted MRI Scans"
subtitle: "A new framework shows how to plug pretrained diffusion models into classical image reconstruction solvers, with injected noise helping iterates escape saddle points."
url: "https://alphasignal.ai/news/uw-madison-s-sgpnp-beats-diffusion-solvers-on-severely-corrupted-mri-scans"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-11-05T06:02:08.921Z"
updated: "2026-11-05T06:05:04.656Z"
type: "Paper"
topics:
  - "Image"
  - "Llms"
subtopics:
  - "Image Generation"
  - "Image Editing"
  - "Neural Rendering"
upvotes: 3412
source_url: "https://arxiv.org/abs/2604.03603"
comments: 0
truncated: true
---

# UW-Madison's SGPnP Beats Diffusion Solvers on Severely Corrupted MRI Scans

> A new framework shows how to plug pretrained diffusion models into classical image reconstruction solvers, with injected noise helping iterates escape saddle points.

Plug-and-play (PnP) methods reconstruct images from incomplete or corrupted measurements by alternating between a data-consistency update and a learned denoiser. Data consistency enforces what the scanner or camera measured; the denoiser supplies an image prior, replacing the proximal operator that a conventional optimizer would use. This pattern supports MRI reconstruction, deblurring, super-resolution, and inpainting without requiring an explicit formula for the prior.

A [new study](https://arxiv.org/abs/2604.03603) from researchers at UW-Madison, Washington University in St. Louis, and Los Alamos National Laboratory finds that conventional PnP methods lose accuracy as measurements become sparse or corruption grows severe. Their Stochastic Generative Plug-and-Play method, or SGPnP, combines a pretrained score-based diffusion model with fresh Gaussian noise at each iteration. It competes with specialized diffusion solvers while preserving the existing PnP optimization loop. The authors have released the [source code](https://github.com/uw-cig/SGPnP).

## Hard corruptions expose PnP’s blind spot

Standard PnP denoisers are usually trained to remove low-level additive Gaussian noise. Intermediate iterates from difficult inverse problems look different: large masks create block-shaped gaps, undersampled MRI produces aliasing, and deblurring leaves structured residuals. This distribution shift gives the denoiser inputs that differ sharply from its training data.

Low-noise denoisers also lack the broad generative behavior needed to infer large missing regions. Re-noising methods such as [SNORE](https://proceedings.mlr.press/v235/renaud24a.html) reduce the distribution mismatch by adding noise before denoising, but their low-noise models provide limited gains on strongly underdetermined tasks. Score-based diffusion models cover a much wider range of corruption levels and learn image structure at multiple scales.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [Stochastic Generative Plug-and-Play Priors](https://arxiv.org/abs/2604.03603)
- [SGPnP GitHub Repository](https://github.com/uw-cig/SGPnP)
- [Plug-and-Play Image Restoration with Stochastic Denoising Regularization (SNORE)](https://proceedings.mlr.press/v235/renaud24a.html)
