MRI Super-Resolution with RCDM/WaveMix and Task-Aware Segmentation

Kavitha Viswanathan; Harsh Choudhary; Amit Sethi

cs.CVSep 30, 2026

MRI Super-Resolution with RCDM/WaveMix and Task-Aware Segmentation

Authors: Kavitha Viswanathan, Harsh Choudhary, Amit Sethi

Organizations: Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai 400 076, India

Abstract

Super-resolution and quality enhancement of 1.5,T brain MRI are normally validated with image-fidelity metrics, although their purpose is to improve downstream analysis. We study whether enhancement improves tissue segmentation, and for which segmenters. We propose an unpaired, physics-guided training pipeline for a lightweight ( $\le$ 2.5,M parameter) recurrent convolutional enhancer: a six-module stochastic 1.5,T degradation operator, a residual adversarial network that adds scanner-specific texture without moving anatomy, and a cycle-consistent objective with an anti-identity penalty that rules out the copy solution. We then train U-Net, Swin-UNet and wavelet token-mixing segmenters \citep{jeevan2023wavemix} from scratch on either raw or enhanced 1.5,T images of the same subjects, using identical labels and subject-level splits, for three enhancer variants and two datasets. On ABIDE (41 held-out subjects, FreeSurfer labels) enhancement significantly improves the wavelet segmenter (mean Dice $+0.014$ , Wilcoxon $p=3.5\times10^{-5}$ ; CSF $+0.018$ , grey matter $+0.013$ ), significantly degrades the U-Net ( $-0.008$ , $p=5.1\times10^{-4}$ ) and leaves Swin-UNet unchanged. On IXI, whose labels come from FSL-FAST, enhancement lowers Dice for all nine pairings, almost entirely through CSF; we trace this to spatially implausible CSF voxels in the labels that penalise smoother predictions. Enhancement of low-field MRI should therefore be validated per downstream model and against reliable labels.

Figures & tables

Module	Parameters	$p$
Anisotropic blur	$\sigma_{xy}\!\in\![0.6,1.8]$ , $\sigma_{z}\!\in\![0.8,2.2]$	0.95
$k$ -space (Rician) noise	$s\in[0.02,0.08]$	0.90
Bias field	strength $\in[0.05,0.20]$	0.70
Intensity remap	$\gamma\in[0.80,1.25]$ , $a\in[0.85,1.15]$	0.50
Image-domain noise	$\sigma\in[0.005,0.025]$	0.30
Partial-volume resampling	$f\in[1.3,1.8]$	0.40

Table 1: Stochastic 1.5 T degradation operator. Each module fires with probability

p

; parameters are drawn uniformly from the ranges.

Panel of Figure 1 — Figure 1: RCDM enhancer adapted for structural brain MRI. Top: a six-module stochastic degradation operator $D_{1.5T}$ and a residual GAN ( $\sim$ 8.4 M params) synthesise 1.5 T-like slices from 3 T volumes. Middle: a group of $S$ adjacent axial slices (GOF = 3 or 5) enters the backbone ( Viswanathan et al., 2025 ) : 2D Haar DWT of the centre slice (beige), 3D deformable alignment (blue), Conv3D residual blocks (pink), then flatten, Conv2D and quality-enhancement head (cyan) gated by a memory tensor (yellow). The output has the same resolution as the input (scale = 1: quality enhancement, not spatial upscaling). Dashed orange: cycle-A re-degradation consistency. Bottom: a frozen 3 T-trained segmentation backbone provides the downstream Dice evaluation.

	PSNR $\uparrow$	SSIM $\uparrow$	MicroSSIM $\uparrow$
Input (epoch 6)	22.37	0.790	0.775
Best (epoch 6)	23.85	0.876	0.859
Final (epoch 21)	23.71	0.861	0.844

Table 2: Validation fidelity on synthetic 1.5 T/3 T pairs (IXI). Best epoch by SSIM.

Figure 2: Synthetic 1.5 T input, enhanced output, 3 T target and absolute error for a held-out slice.

	Metric	RAW	SR	$\Delta$ ( $k$ )	$p$
U-Net	mDice	.867	.860	$\mathbf{-.008}$ (9)	$\mathbf{5.1\text{e-}4}$
	CSF	.898	.891	$-.007$ (10)	$4.7\text{e-}3$
	GM	.838	.837	$-.001$ (18)	.28
	WM	.867	.851	$\mathbf{-.016}$ (10)	$\mathbf{2.0\text{e-}3}$
Swin	mDice	.874	.878	$+.004$ (22)	.32
	CSF	.907	.903	$-.004$ (16)	.27

Table 3: ABIDE segmentation (GOF = 5 enhancer,

n{=}41

\Delta

: mean paired difference;

k

: subjects improved;

p

: Wilcoxon. Bold: significant after Bonferroni (

\times 12

Figure 3: ABIDE paired

\Delta

Dice (SR

-

RAW), mean

\pm

95% CI over 41 subjects. Stars: Bonferroni-corrected Wilcoxon

p

<

0.05, **

<

0.01, ***

<

0.001).

Figure 4: Change in mean Dice for every enhancer variant $\times$ segmentation backbone. Left: ABIDE (FreeSurfer labels); stars mark significant paired Wilcoxon tests. Right: IXI (FSL-FAST labels): all pairings lose.

Figure 5: ABIDE example (axial). From left: real 1.5 T, enhanced, label, RAW prediction, SR prediction (blue CSF, green GM, orange WM).

Figure 6: IXI example. From left: real 1.5 T, enhanced, FSL-FAST label, RAW prediction, SR prediction. Isolated CSF voxels (blue) inside tissue in the label are not visible in the image.

Figure 7: A second IXI slice with the same layout; the SR prediction is smoother than the label.

Explore similar work

May 27, 2026cs.CV

Enhancing Ultra-low-field MRI with Segmentation-guided Adversarial Learning

Ultra-low-field (ULF) MRI offers portable and low-cost imaging but suffers from poor image quality. To address this, we present our submission to the 2025 ULF Enhancement Challenge (ULF-EnC), where the goal is to synthesise high-field-like MRIs from 64 mT scans. Our pipeline enhances ULF MRI through a combination of anatomical conditioning and model ensembling. We first generate tissue segmentation priors using a Swin UNETR trained solely on challenge-provided data. These priors condition two independent enhancement networks - a CycleGAN and a transformer-based residual enhancement model (T-REX) - each trained to synthesise 3 T-like MRIs. Outputs from both models are combined using a weighted average. Our approach produces enhanced MRIs that were comparable to high-field scans both quantitatively and qualitatively.

James Grover, Andrew Phair, Michael Ferraro +1

Image X Institute, Sydney School of Health Sciences, Faculty of Medicine and Health, The University of Sydney

Magnetic Resonance Imaging U-Net Transformer Residual Streams Adversarial Training

Ask PDF

May 15, 2026eess.IV

Layer Selection in Feature-Based Losses Affects Image Quality and Microstructural Consistency in Deep Learning Super-Resolution of Brain Diffusion MRI

Clinical application of high-resolution diffusion MRI is hindered by hardware limitations and prohibitive scan times, motivating computational super-resolution. This study investigates the efficacy of a feature-based loss function in preserving diffusion signal consistency in deep learning super-resolution. Using 7T data from the human connectome project to generate pairs of low- and high-resolution diffusion weighted images (DWI), we trained UNets for 2D super-resolution. Ablation and isolation studies evaluated different VGG16-layers for feature-based losses against an image-based L1 baseline. Deeper layers and combinations thereof resulted in grid-like artifacts in super-resolution DWIs, which persisted in diffusion parameters like quantitative and fractional anisotropy. No such artifacts were present when using the shallowest layer. Downstream analysis for this layer showed great consistency with the ground truth, even for 9-fold super-resolution. Image SNR and used VGG16-layer depths modulated artifact appearance and severity, mandating careful selection of contributing layers for application in and beyond diffusion MRI.

David Lohr, Rene Werner

Institue for Applied Medical Informatics, University Medical Center Hamburg-Eppendorf, Hamburg, 20246, Germany · Institute of Computational Neuroscience, University Medical Center Hamburg-Eppendorf, 20246, Hamburg, Germany · Center for Biomedical Artificial Intelligence (bAIome), University Medical Center Hamburg-Eppendorf, 20246, Hamburg, Germany

Generative Super-Resolution Magnetic Resonance Imaging Residual Conditional Diffusion Model Loss Function Image Quality Assessment Porous Media

Ask PDF

Aug 9, 2026cs.CV

MRI super-resolution in ten sampling steps using a diffusion bridge model

Objective. MRI provides excellent soft-tissue contrast, but long acquisition times can cause patient discomfort and lead to motion artifacts, forcing a trade-off between spatial resolution and scan time. Diffusion-based super-resolution (SR) reconstructs high-resolution (HR) images from low-resolution (LR) inputs, but typically needs many sampling steps and initializes from a Gaussian prior ill-suited to image restoration. We developed an efficient diffusion framework that reconstructs HR MRI directly from LR data. Approach. We propose super-resolution diffusion bridge model (SR-DBM), a super-resolution diffusion bridge model that casts SR as a stochastic transport between the LR and HR image distributions. Through a Doob's h-transform of a mean-reverting stochastic differential equation, SR-DBM pins the process to the paired HR and LR images at its endpoints, initializing reconstruction from the measured anatomy rather than from Gaussian noise. The HR image is recovered by a deterministic reverse trajectory in which a network predicts the clean image at each of only ten sampling steps. We evaluated SR-DBM on ultra-high-field 7T brain T1 MP2RAGE maps and pelvic T2-weighted prostate images against nine comparison methods using PSNR, SSIM, GMSD, and LPIPS. Main results. SR-DBM attained the highest PSNR and SSIM and the lowest GMSD on both datasets (brain: 27.66+-1.52 dB, 0.96+-0.02, 7.96+-1.86$; prostate: 27.87+-2.29 dB, 0.80+-0.05, 8.38+- 1.44), with statistically significant gains over every comparison method (two-sided Wilcoxon signed-rank test with Holm correction, p<0.05). The strongest baseline, SR-EMamba, ranked second. Qualitatively, SR-DBM produced the smallest residual errors and best preserved fine structures and lesions.

Mojtaba Safari, Hang Yu, Zach Eidex +10

Department of Radiation and Cellular Oncology, The University of Chicago, Chicago, IL, United States · Department of Physics and Astronomy, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States · Department of Radiation Oncology and Winship Cancer Institute, Emory University, Atlanta, GA, United States +3

Generative Super-Resolution Magnetic Resonance Imaging Diffusion Bridges Signal-To-Noise Ratio

Ask PDF