Research map / Image generation / Image restoration
Rethinking Real-World MRI Denoising: Learning from Physical Noise
Sebastian Rassmann, David Kügler, Sascha Brunheim, Philipp Ehses, Martin Reuter
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with diffusion re-noised pairs (ReN2N) instead of physical repeats (pN2N) | uYADO · HCP-E test, T1w | SSIM (%): 89.65 -> 90.60 | Improved |
| train with ReN2N instead of pN2N | uYADO · OASIS-3 test, T1w | SSIM (%): 89.39 -> 89.41 | No clear effect |
| train with ReN2N instead of pN2N, internal T2w test set | uYADO · Rhineland Study test, T2w | SSIM (%): 88.80 -> 88.58 | Hurt |
| add co-acquired T2w guidance | YADO (pN2N) · OASIS-3 test, T1w | SSIM (%): 89.39 -> 90.03 | Improved |
| add anisotropic T2w guidance | YADO (pN2N) · Kirby, T1w | SSIM (%): 95.69 -> (94.63) | Hurt |
| add guidance from anisotropic T2w on motion FLAIR | YADO (pN2N) · FLAIR still | SSIM (%): 84.89 -> (69.94) | Hurt |
| replace degraded input with physical re-acquisition (pNer2N) (ablation) | Noisier2Noise · OASIS-3 validation | SSIM (%): 90.13 -> 90.25 | Improved |
| replace blind-spot input with physical re-acquisition (pS-N2Void) (ablation) | Structured N2Void · OASIS-3 validation | SSIM (%): 90.35 -> 90.73 | Improved |
| use diffusion-model re-noising instead of trivial Rician re-noising (ablation) | ReN2N · OASIS-3 validation | SSIM (%): 89.52 -> 91.00 | Improved |
| add Gaussian blur to trivial Rician re-noising (ablation) | riReN2N · OASIS-3 validation | SSIM (%): 89.52 -> 90.35 | Improved |
| add ReN augmentation to baseline training (ablation) | Noisier2Noise · OASIS-3 test, T1w | SSIM (%): 88.59 -> 88.79 | Improved |
| add ReN augmentation to baseline training (ablation) | Structured N2Void · OASIS-3 test, T1w | SSIM (%): 88.69 -> 88.65 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Neighbor2Neighbor — Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images (CVPR 2021) · better on 1 of 1 shared result
- Noise2Void — Noise2Void - Learning Denoising From Single Noisy Images (CVPR 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Score-based Self-supervised MRI Denoising (ICLR 2025 · similarity 0.89)
- Self-Supervised Diffusion MRI Denoising via Iterative and Stable Refinement (ICLR 2025 · similarity 0.88)
- Efficient Noise Calculation in Deep Learning-based MRI Reconstructions (ICML 2025 · similarity 0.87)
- Patch2Self2: Self-supervised Denoising on Coresets via Matrix Sketching (CVPR 2024 · similarity 0.87)
- Resfusion: Denoising Diffusion Probabilistic Models for Image Restoration Based on Prior Residual Noise (NeurIPS 2024 · similarity 0.86)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
