ReDiff: Reliability-Guided Diffusion for Trustworthy Ultra-Low-Field to High-Field MRI Synthesis

Zhenxuan Zhang; Peiyuan Jing; Ruicheng Yuan; Liwei Hu; Anbang Wang; Fanwen Wang; Yinzhe Wu; Kh Tohidul Islam

cs.CVMar 11, 2026

ReDiff: Reliability-Guided Diffusion for Trustworthy Ultra-Low-Field to High-Field MRI Synthesis

Authors: Zhenxuan Zhang, Peiyuan Jing, Ruicheng Yuan, Liwei Hu, Anbang Wang, Fanwen Wang, Yinzhe Wu, Kh Tohidul Islam, +4 more

Organizations: Department of Bioengineering and Imperial-X, Imperial College London, London, UK · Department of Electrical and Computer Systems Engineering, Monash University, Melbourne, Australia · National Heart and Lung Institute, Imperial College London, London SW7 2AZ, UK · Cardiovascular Research Centre, Royal Brompton Hospital, London SW3 6NP, UK · School of Biomedical Engineering & Imaging Sciences, King’s College London, London WC2R 2LS, UK

Abstract

Low-field to high-field MRI synthesis has emerged as a promising strategy to improve image quality when access to high-field scanners is limited. However, in ultra-low-field settings, the degradation of anatomical detail is spatially heterogeneous: structurally ambiguous regions are more susceptible to unstable high-frequency generation, which may produce anatomically inconsistent textures and boundaries. This issue is particularly problematic when synthesized images are used for downstream quantitative analysis. We therefore study how to make diffusion-based LF-to-HF synthesis more spatially reliable, rather than only sharper on average. To this end, we propose a reliability-guided diffusion framework (ReDiff) with two complementary inference-time mechanisms. First, a reliability-guided sampling strategy attenuates unstable reverse-diffusion updates in regions with weak low-field support. Second, an uncertainty-aware candidate selection scheme aggregates multiple stochastic reconstructions according to spatial consensus and predictive uncertainty. Beyond aggregate image quality, we test whether the uncertainty is itself a usable reliability signal. Experiments on paired 64mT $\rightarrow$ 3T MRI datasets show that ReDiff attains the lowest LPIPS across three contrasts and two datasets while remaining competitive on PSNR and SSIM, and downstream segmentation analysis indicates better preservation of anatomical structure.

Figures & tables

Figure 1: Motivation and challenges of our Reliability-Guided Diffusion. (A) Low-field MRI (e.g., 64 mT) is cost-effective and accessible but typically produces lower-quality images. LF-to-HF synthesis seeks to recover high-field (e.g., 3T) image fidelity under hardware and acquisition constraints. (B) Despite recent progress, diffusion-based methods may introduce unreliable high-resolution artifacts, such as artificial textures and spurious edges. These errors can degrade downstream tasks, leading to inaccurate segmentation, biased volume estimation, and reduced clinical trust.

Figure 2: Reliability-Guided Diffusion (ReDiff) framework for LF-to-HF MRI synthesis. Given a low-field (LF) image $x$ , a conditional diffusion U-Net predicts the noise term $\epsilon_{\theta}(y_{t},x,t)$ to iteratively recover the high-field (HF) image. (a) In the proposed reliability-guided sampling (RGS), a timestep-dependent reliability map $R_{t}(x,y_{t})$ modulates the reverse diffusion update to suppress unsupported high-frequency amplification in ill-posed regions. (b) To further enhance robustness, uncertainty-aware candidate selection (UCS) generates multiple candidates and performs reliability-weighted aggregation based on spatial uncertainty. Together, the two-stage design improves structural consistency while reducing hallucinated details.

Figure 3: Qualitative comparison of multi-contrast low-field to high-field MRI synthesis. For each case, the top row shows full images and the bottom row shows zoomed views of the red boxed regions. ReDiff produces sharper anatomical structures and fewer artifacts, with PSNR and SSIM reported on each result.

Setting	Method	Private Dataset			Leiden Uni. Dataset
Setting	Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
T 1 w	Low-field	24.83 $\pm$ 2.77 ∗∗	0.758 $\pm$ 0.133 ∗∗	0.2222 $\pm$ 0.0866 ∗∗	24.73 $\pm$ 2.81 ∗∗	0.768 $\pm$ 0.117 ∗∗	0.2138 $\pm$ 0.0741 ∗∗
	Pix2Pix ( Isola et al. 2017 )	30.51 $\pm$ 2.62 ∗∗	0.933 $\pm$ 0.052 ∗∗	0.1050 $\pm$ 0.0394 ∗∗	30.71 $\pm$ 2.64 ∗∗	0.939 $\pm$ 0.054 ∗∗	0.1043 $\pm$ 0.0366 ∗∗
	ESRGAN ( Wang et al. 2018 )	31.72 $\pm$ 2.54 ∗∗	0.949 $\pm$ 0.038 ∗	0.1080 $\pm$ 0.0746 ∗∗	32.08 $\pm$ 2.46 ∗∗	0.958 $\pm$ 0.029	0.1034 $\pm$ 0.0704 ∗∗
	TransUNet ( Chen et al. 2024 )	31.71 $\pm$ 2.75 ∗∗	0.939 $\pm$ 0.092 ∗∗	0.0947 $\pm$ 0.0720 ∗	32.08 $\pm$ 2.72 ∗∗	0.950 $\pm$ 0.069 ∗	0.0901 $\pm$ 0.0592 ∗
	ResViT ( Dalmaz, Yurt, and Cukur 2022 )	31.61 $\pm$ 2.75 ∗∗	0.941 $\pm$ 0.058 ∗∗	0.0887 $\pm$ 0.0505 ∗	32.10 $\pm$ 2.65 ∗∗	0.953 $\pm$ 0.045 ∗	0.0838 $\pm$ 0.0435

Table 1: Evaluation on the paired 64mT

\rightarrow

3T dataset under Multi-Contrast setting. ∗ indicates

p<0.05

and ∗∗ indicates

p<0.01

in a paired Wilcoxon signed-rank test against our ReDiff model, computed over test slices; absence of a marker indicates that the difference from ReDiff is not significant. Bold indicates the best result and underline indicates the second best.

Diff	RGS	UCS	Private Dataset
Diff	RGS	UCS	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
$\checkmark$	$\times$	$\times$	30.98 $\pm$ 2.55	0.934 $\pm$ 0.060	0.1080 $\pm$ 0.0697
$\checkmark$	$\checkmark$	$\times$	31.72 $\pm$ 2.51	0.945 $\pm$ 0.052	0.1062 $\pm$ 0.0717
$\checkmark$	$\times$	$\checkmark$	31.45 $\pm$ 2.60	0.940 $\pm$ 0.056	0.0991 $\pm$ 0.0687
$\checkmark$	$\checkmark$	$\checkmark$	33.00 $\pm$ 2.66	0.959 $\pm$ 0.038	0.0838 $\pm$ 0.0595
Diff	RGS	UCS	Leiden Uni. Dataset

Table 2: Ablation study of the proposed ReDiff framework on the Private and Leiden datasets, averaged over the T 1 w, T 2 w, and FLAIR contrasts. RGS denotes reliability-guided sampling and UCS denotes uncertainty-aware candidate selection. Bold indicates the full model.

Figure 4: UCS uncertainty as a reliability signal. (a) Slice-level validation. Left : mean UCS uncertainty vs. slice MAE, coloured by PSNR (markers: octile means, 95% CI). Centre : MAE per uncertainty decile (shaded, 95% CI) against an affine recalibration (dotted). Right : ROC for identifying bottom-decile slices (PSNR $\leq 27.89$ dB), using mean slice uncertainty as the score. (b) Representative slices for four diagnostic regimes; $|$ error $|$ and UCS share a colour scale across rows, and the orange box marks the worst error window, magnified at right.

UCS	$\bar{u}$	PSNR [dB]	SSIM	Edge MAE	Failure [%]
Q1	0.0043	33.55 $\pm$ 3.16	0.9256 $\pm$ 0.1538	0.0641	4.2 [2.8, 6.3]
Q2	0.0064	32.52 $\pm$ 2.29	0.9525 $\pm$ 0.0740	0.0702	4.1 [2.7, 6.1]
Q3	0.0079	31.63 $\pm$ 2.33	0.9573 $\pm$ 0.0386	0.0755	6.5 [4.7, 8.8]
Q4	0.0098	30.59 $\pm$ 2.50	0.9486 $\pm$ 0.0793	0.0810	10.7 [8.4, 13.6]
Q5	0.0136	29.61 $\pm$ 2.84	0.9382 $\pm$ 0.0879	0.0884	24.7 [21.2, 28.5]

Table 3: Fidelity and failure rate stratified by UCS uncertainty level. Values are mean

\pm

s.d.; failure rate is reported with its Wilson 95% confidence interval.

\bar{u}

is the mean UCS uncertainty within the stratum.

Figure 5: Qualitative and quantitative downstream validation. (a) 3D structural visualization of SynthSeg-derived segmentations for representative cases. (b) Volumetric correlation with respect to the 3T reference.

Figure 6: Inference-efficiency analysis of ReDiff. (a) Component overhead measured as runtime per volume under different inference configurations. (b) Sampling-step scaling analysis. Runtime decreases substantially as the number of diffusion sampling steps is reduced.

Data
Setting	Value
Paired dataset	Monash / Leiden, aligned
Case list	shared, identical across all methods
Case selection	all
Input / output channels	1 / 1
Resolution	224 $\times$ 224

Table S1: Configuration of the model and of the evaluation protocol used for every number in this appendix.

Method	PSNR [dB]		SSIM		LPIPS
	Mean [95% CI]	Median [IQR]	Mean [95% CI]	Median [IQR]	Mean [95% CI]	Median [IQR]
Low-field input	23.27 [23.17, 23.37]	23.35 [21.44, 25.17]	0.6793 [0.6734, 0.6853]	0.7065 [0.6036, 0.7947]	0.2310 [0.2280, 0.2342]	0.2175 [0.1765, 0.2706]
Pix2Pix	29.31 [29.22, 29.40]	29.03 [27.58, 30.53]	0.9202 [0.9161, 0.9240]	0.9427 [0.9206, 0.9568]	0.1091 [0.1074, 0.1110]	0.1029 [0.0849, 0.1250]
ESRGAN	29.63 [29.50, 29.76]	29.96 [27.58, 32.14]	0.9191 [0.9147, 0.9234]	0.9504 [0.9208, 0.9668]	0.1774 [0.1740, 0.1809]	0.1572 [0.1036, 0.2449]
ResViT	31.14 [31.04, 31.25]	31.16 [29.40, 32.79]	0.9392 [0.9354, 0.9428]	0.9623 [0.9445, 0.9738]	0.0911 [0.0888, 0.0934]	0.0787 [0.0616, 0.1027]
CyTran	32.29 [32.17, 32.41]	32.36 [30.52, 34.19]	0.9516 [0.9480, 0.9550]	0.9720 [0.9573, 0.9808]	0.0933 [0.0914, 0.0952]	0.0838 [0.0642, 0.1113]

Table S2: Comparison on the shared paired case list, identical for every method; three cases produced degenerate PSNR values for CyTran and are excluded from that entry alone. Each image is min-max normalised to

[0,1]

independently, including the reference; PSNR uses peak

=2

; SSIM is a single global window. Confidence intervals are percentile bootstrap over 10000 resamples. Best mean in bold , second best underlined .

Figure S1: Per-case metric distributions across the shared case list for (a) PSNR, (b) SSIM and (c) LPIPS. The violin shows the distribution, the box the interquartile range, the white line the median and the diamond the mean with its bootstrap

95\%

confidence interval. Note the long inferior PSNR tail of ESRGAN and the bimodal LPIPS distributions of ESRGAN and MiDiffusion, neither of which is apparent from summary statistics.

Metric	Baseline	$\Delta$ [95% CI]	$d_{z}$	$p$
PSNR [dB], ReDiff minus baseline
	Low-field	9.03 [8.93, 9.13]	3.43	< 0.001
	Pix2Pix	2.99 [2.91, 3.07]	1.40	< 0.001
	ESRGAN	2.67 [2.59, 2.75]	1.31	< 0.001
	ResViT	1.16 [1.11, 1.22]	0.78	< 0.001
	CyTran	0.02 [-0.05, 0.08]	0.01	0.512

Table S3: Paired comparisons of ReDiff against each baseline over the shared case list. Every row uses the same cases, with the single exception of the CyTran PSNR row, from which three degenerate cases are excluded.

\Delta

is the within-case difference, intervals are percentile bootstrap over 10000 resamples,

d_{z}

is the paired Cohen effect size, and

p

is a two-sided Wilcoxon signed-rank test with Holm–Bonferroni correction across every row of the table.

Figure S2: Within-case differences between ReDiff and each baseline for (a) PSNR, (b) SSIM and (c) LPIPS. Each dot is one case and the red bar is the bootstrap

95\%

confidence interval of the mean difference. Positive is better for PSNR and SSIM, negative is better for LPIPS. The CyTran column is centred on zero with wide symmetric scatter on the fidelity metrics and shifted negative on LPIPS.

Figure S3: Gain analysis, one row per metric: PSNR (a–c), SSIM (d–f) and LPIPS (g–i). Left : per-case score against the score of its own low-field input, with the identity line; every case falls on the improved side. Centre : gain over the input against input quality, with a least-squares fit; the negative slope shows the gain concentrating on the worst inputs. Right : win rate against each baseline, with normal-approximation $95\%$ intervals.

Metric	Stage	Mean [95% CI]	SD	Median [IQR]
PSNR	64 mT input	23.46 [22.95, 23.98]	2.64	23.49 [21.75, 25.42]
	Base(A)	24.50 [23.98, 25.04]	2.76	24.30 [22.75, 26.53]
	ReDiff	28.00 [27.51, 28.48]	2.54	27.95 [26.53, 29.99]
SSIM	64 mT input	0.680 [0.645, 0.712]	0.177	0.721 [0.599, 0.801]
	Base(A)	0.790 [0.756, 0.820]	0.170	0.830 [0.758, 0.897]
	ReDiff	0.875 [0.844, 0.901]	0.152	0.927 [0.870, 0.948]

Table S4: Stage-wise reconstruction quality on

n=105

paired validation samples, isolating the contribution of the sampled residual over the deterministic base network. Protocol as in Table S1 . Intervals are percentile bootstrap over 10000 resamples and comparisons are paired within sample.

Figure S4: Decile-resolution extension of the uncertainty stratification. (a) Slice MAE against UCS decile, Spearman

\rho=0.59

. (b) Slice PSNR against UCS decile,

\rho=-0.49

. (c) Failure rate per decile, with a ratio of

4.6\times

between D10 and D1. The failure rate is flat across the middle deciles and rises steeply over the top three, so the signal acts as an alarm rather than a graded risk estimate.

Decile	$n$	mean UCS	PSNR [dB]	MAE	Failure [%]
D1	272	0.0036	33.93	0.0183	6.2
D2	271	0.0051	33.17	0.0196	2.2
D3	271	0.0060	32.78	0.0212	3.7
D4	271	0.0067	32.26	0.0233	4.4
D5	271	0.0074	32.13	0.0240	4.1
D6	271	0.0083	31.12	0.0280	8.9

Table S5: Fidelity and failure rate across ten UCS uncertainty deciles, extending Table 3 of the main text to finer resolution. A slice counts as a failure if its PSNR is at most

27.89

dB. Source: comprehensive_notebook , not re-scored for this appendix.

Figure S5: Rank-selected qualitative comparison. Rows are the worst, median and best case by PSNR on the shared case list, so the selection is reproducible rather than curated. Columns are the low-field input, five baselines, ReDiff and the paired $3$ T reference, with PSNR, SSIM and LPIPS printed per panel. Every method fails on the worst case; all methods converge on the best case.

Figure S6: Diagnostic gallery across five regimes, one per row: low uncertainty with high fidelity, a typical median case, high uncertainty with high error, low uncertainty with high error, and the case with the largest residual shift. Columns are the low-field input, the base network output, ReDiff, the $3$ T reference, the absolute error and the UCS uncertainty map. Error and uncertainty share a colour scale within each row. The fourth row is a silent failure and is included to show the limit of the signal.

Explore similar work

ULF-Synth: Physics-Guided Ultra-Low-Field MRI Enhancement for Pediatric Neuroimaging
May 23, 2026Toufiq Musah, Salvatore Calcagno, Federica Proietto Salanitri +3Medical Image Super-Resolution Magnetic Resonance Imaging
Kwame Nkrumah University of Science and Technology, Kumasi, Ghana · University of Catania, Catania, Italy · The Hong Kong University of Science and Technology, Hong Kong SAR, China +1
Uncertainty-Aware 3D Residual Wavelet Diffusion for Ultra Low-Field MRI Super-Resolution
Sep 21, 2026Rui W. Yeow, Millie Beament, Fred Dick +6Medical Image Super-Resolution Magnetic Resonance Imaging
University College London, London, UK
Subject-Specific Low-Field MRI Synthesis via a Neural Operator
Mar 26, 2026Ziqi Gao, Nicha Dvornek, Xiaoran Zhang +3Magnetic Resonance Imaging Medical Image Synthesis
Departments of Biomedical Engineering, Yale University · Department of Radiology & Biomedical Imaging, Yale University

May 23, 2026J/K move · Enter open · S save

ULF-Synth: Physics-Guided Ultra-Low-Field MRI Enhancement for Pediatric Neuroimaging

Toufiq Musah, Salvatore Calcagno, Federica Proietto Salanitri, Xiaomeng Li+2

Kwame Nkrumah University of Science and Technology, Kumasi, Ghana · University of Catania, Catania, Italy · The Hong Kong University of Science and Technology, Hong Kong SAR, China+1

Medical Image Super-Resolution Magnetic Resonance Imaging Medical Image Synthesis Neuroimaging

Open paper AskPDF arXiv

Ultra-low-field (ULF) MRI offers portable and accessible neuroimaging but suffers from reduced signal-to-noise ratio and limited spatial resolution compared to high-field (HF) systems. Acquiring paired ULF-HF data for supervised enhancement is often difficult, particularly in resource-limited settings. We introduce ULF-Synth, a framework that combines: (i) acquisition-based synthesis of realistic ULF images from HF volumes to create large-scale paired training data, (ii) a spatial-frequency domain objective that prioritizes recovery of high-frequency anatomical detail. This formulation is architecture-agnostic, consistently improving structural similarity and perceptual fidelity across encoder-decoder, adversarial, and diffusion-based translation models. When trained exclusively on synthetic data, the resulting models generalize effectively to real 64mT ULF acquisitions, improving downstream multiclass brain segmentation and achieving higher radiologist preference and diagnostic acceptability in a blinded reader study. These findings demonstrate that synthetic paired supervision provides a practical and scalable pathway for enhancing ULF MRI without requiring real paired acquisitions. Code, Models and Dataset: https://github.com/toufiqmusah/ULF-Synth

Abstract

Explore similar work

ULF-Synth: Physics-Guided Ultra-Low-Field MRI Enhancement for Pediatric Neuroimaging

Uncertainty-Aware 3D Residual Wavelet Diffusion for Ultra Low-Field MRI Super-Resolution

Subject-Specific Low-Field MRI Synthesis via a Neural Operator