Comparative Analysis of 3D Convolutional and 2.5D Slice-Conditioned U-Net Architectures for MRI Super-Resolution via Elucidated Diffusion Models

Hendrik Chiche; Ludovic Corcos; Logan Rouge

cs.CVMar 15, 2026

Comparative Analysis of 3D Convolutional and 2.5D Slice-Conditioned U-Net Architectures for MRI Super-Resolution via Elucidated Diffusion Models

Authors: Hendrik Chiche, Ludovic Corcos, Logan Rouge

Organizations: GENCI (Grand Équipement National de Calcul Intensif), France

Abstract

Magnetic resonance imaging (MRI) super-resolution (SR) methods that computationally enhance low-resolution acquisitions to approximate high-resolution quality offer a compelling alternative to expensive high-field scanners. In this work we investigate an elucidated diffusion model (EDM) framework for brain MRI SR and compare two U-Net backbone architectures: (i) a full 3D convolutional U-Net that processes volumetric patches with 3D convolutions and multi-head self-attention, and (ii) a 2.5D slice-conditioned U-Net that super-resolves each slice independently while conditioning on an adjacent slice for inter-slice context. Both models employ continuous-sigma noise conditioning following Karras et al. and are trained on the NKI cohort of the FOMO60K dataset. On a held-out test set of 5 subjects (6 volumes, 993 slices), the 3D model achieves 37.75 dB PSNR, 0.997 SSIM, and 0.020 LPIPS, improving on the off-the-shelf pretrained EDSR baseline (35.57 dB / 0.024 LPIPS) and the 2.5D variant (35.82 dB) across all three metrics under the same test data and degradation pipeline.

Figures & tables

Fig. 1: Overview of the two super-resolution pipelines. (a) The 3D model (best-performing) extracts volumetric patches, processes them through a 3D U-Net with 20-step Euler sampling, and blends overlapping patches to reconstruct the full HR volume. (b) The 2.5D model extracts individual slices with one neighboring slice as context, processes each through a 2D U-Net with one-step Heun sampling, and stacks results.

Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Params
Bicubic interpolation	33.89	0.957	0.091	—
EDSR [ 10 ] (DIV2K)	35.57	0.977	0.024	1.4 M
Swin2SR [ 22 ] (DIV2K)	35.50	0.978	0.024	1.0 M
2.5D EDM (ours, 10 ep)	35.82	0.971	0.040	51.1 M
3D EDM (ours, 20 ep)	37.75	0.997	0.020	50.7 M

TABLE I: Quantitative results for

2\times

MRI super-resolution on the NKI test set (5 subjects, 993 sagittal slices / 6 volumes). All methods evaluated on identical test data and degradation pipeline. EDSR and Swin2SR use pretrained DIV2K weights without MRI fine-tuning.

Fig. 2: PSNR and SSIM comparison across methods for

2\times

MRI SR on the NKI test set. The 3D EDM model achieves the strongest PSNR and SSIM among the evaluated methods.

Fig. 3: Visual comparison of all methods on a mid-sagittal slice (sub_2642). Top row: full slice. Bottom row: zoomed region of interest (red box) showing cortical folds and gray/white matter boundaries. The 3D EDM model recovers the sharpest anatomical detail, while EDSR and Swin2SR produce smoother outputs. Bicubic interpolation exhibits visible blurring.

	3D EDM (ours)			2.5D EDM (ours)			Bicubic
Subject	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS
sub_2642	37.44	0.996	0.022	35.52	0.968	0.043	33.59	0.954	0.097
sub_2952	38.15	0.997	0.019	36.79	0.971	0.038	35.34	0.956	0.089
sub_6794 (ses1)	37.33	0.997	0.020	34.64	0.971	0.040	32.19	0.958	0.095
sub_6794 (ses2)	37.74	0.997	0.020	35.41	0.972	0.041	33.32	0.959	0.093
sub_8390	38.10	0.996	0.020	35.97	0.970	0.038	33.87	0.956	0.090

TABLE II: Per-subject evaluation on the NKI test set (5 volumes). All methods evaluated on identical test data and degradation pipeline.

Fig. 4: Per-slice PSNR across the sagittal axis for subject sub_2642 (2.5D model).

Fig. 5: 3D EDM model visual comparison across sagittal, axial, and coronal views. Each panel shows (left to right): ground truth HR, 3D model prediction, and trilinear baseline. The 3D model recovers fine cortical detail and inter-slice continuity lost in interpolation.

Fig. 6: Per-pixel absolute error heatmaps for a mid-sagittal slice. Darker regions indicate lower reconstruction error. The EDM models show reduced error in cortical folds and gray/white matter boundaries compared to bicubic and trilinear interpolation; the 3D model achieves the lowest overall error (see Table I ).

Explore similar work

MRI super-resolution in ten sampling steps using a diffusion bridge model
Aug 9, 2026Mojtaba Safari, Hang Yu, Zach Eidex +10Medical Image Super-Resolution Diffusion-Based Inverse Problems
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
Layer Selection in Feature-Based Losses Affects Image Quality and Microstructural Consistency in Deep Learning Super-Resolution of Brain Diffusion MRI
May 15, 2026David Lohr, Rene WernerPerceptual Loss Medical Image Super-Resolution
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
MRI Super-Resolution with RCDM/WaveMix and Task-Aware Segmentation
Sep 30, 2026Kavitha Viswanathan, Harsh Choudhary, Amit SethiMedical Image Analysis Medical Image Super-Resolution
Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai 400 076, India

Aug 9, 2026J/K move · Enter open · S save

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

Mojtaba Safari, Hang Yu, Zach Eidex, Mingzhe Hu+9

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

Medical Image Super-Resolution Diffusion-Based Inverse Problems Diffusion MRI Diffusion Model Acceleration Brain MRI Diffusion Models

Open paper AskPDF arXiv

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.

Abstract

Explore similar work

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

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

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