Unrolled Reconstruction with Integrated Super-Resolution for Accelerated 3D LGE MRI

Md Hasibul Husain Hisham; Shireen Elhabian; Ganesh Adluru; Jason Mendes; Andrew Arai; Eugene Kholmovski; Ravi Ranjan; Edward DiBella

cs.CVMar 18, 2026

Unrolled Reconstruction with Integrated Super-Resolution for Accelerated 3D LGE MRI

Authors: Md Hasibul Husain Hisham, Shireen Elhabian, Ganesh Adluru, Jason Mendes, Andrew Arai, Eugene Kholmovski, Ravi Ranjan, Edward DiBella

Organizations: Kahlert School of Computing, University of Utah, Salt Lake City, UT, USA · Radiology & Imaging Sciences, University of Utah, Salt Lake City, UT, USA · Cardiology, University of Utah, Salt Lake City, UT, USA

Abstract

Accelerated 3D late gadolinium enhancement (LGE) MRI requires robust reconstruction methods to recover thin atrial structures from undersampled k-space data. While unrolled model-based networks effectively integrate physics-driven data consistency with learned priors, they operate at the acquired resolution and may fail to fully recover high-frequency detail. We propose a hybrid unrolled reconstruction framework in which an Enhanced Deep Super-Resolution (EDSR) network replaces the proximal operator within each iteration of the optimization loop, enabling joint super-resolution enhancement and data consistency enforcement. The model is trained end-to-end on retrospectively undersampled preclinical 3D LGE datasets and compared against compressed sensing, Model-Based Deep Learning (MoDL), and self-guided Deep Image Prior (DIP) baselines. Across acceleration factors, the proposed method consistently improves PSNR and SSIM over standard unrolled reconstruction and better preserves fine cardiac structures, leading to improved LA (left atrium) segmentation performance. These results demonstrate that integrating super-resolution priors directly within model-based reconstruction provides measurable gains in accelerated 3D LGE MRI.

Figures & tables

Figure 1: Proposed Unrolled EDSR framework. The EDSR Net serves as the learned proximal operator at each of the $N$ unrolled iterations, with weights $\phi$ shared across iterations. The initial reconstruction $\mathbf{x}^{(0)}=\mathbf{A}^{H}\mathbf{y}$ is the zero-filled adjoint, and data consistency (DC) enforces k-space fidelity via conjugate gradient at each step.

Figure 2: Qualitative comparison of reconstruction methods at acceleration factors $R=4$ and $R=6$ . Each column shows a representative axial slice from a preclinical subject. At $R=4$ , Unrolled+EDSR and Unrolled both recover fine cardiac structures faithfully, while Compressed Sensing introduces visible artifacts and DIP shows smoothing. At $R=6$ , quality degrades across all methods, with Unrolled+EDSR best preserving structural detail. The LA is zoomed in for better visibility.

Method	$R=4$		$R=6$
Method	PSNR (dB)	SSIM	PSNR (dB)	SSIM
Compressed Sensing	$31.6\pm 1.0$	$0.865\pm 0.019$	$28.6\pm 1.4$	$0.824\pm 0.020$
Self-Guided DIP	$32.8\pm 1.2$	$0.894\pm 0.018$	$29.3\pm 1.1$	$0.855\pm 0.015$
Unrolled	$35.1\pm 1.2$	$0.906\pm 0.015$	$32.4\pm 1.5$	$0.878\pm 0.012$
EDSR-Unrolled	$\mathbf{35.6\pm 1.1}$	$\mathbf{0.915\pm 0.014}$	$\mathbf{32.9\pm 1.6}$	$\mathbf{0.889\pm 0.015}$

Table 1: Quantitative comparison of reconstruction methods averaged across 5 preclinical subjects at acceleration factors

R=4

and

R=6

. Best results in bold .

Method	DSC
Compressed Sensing	$0.809$
Self-Guided DIP	$0.835$
Unrolled	$0.884$
Unrolled EDSR (Ours)	$\mathbf{0.893}$
Ground Truth (upper bound)	$0.928$

Table 2: LA segmentation DSC on reconstructed test slices. Results at R = 4.

$N$	PSNR (dB)	SSIM	GPU Memory (GB)
5	$34.2\pm 1.1$	$0.897\pm 0.012$	16.3
7	$35.6\pm 1.1$	$0.915\pm 0.012$	17.4
9	$35.9\pm 1.1$	$0.923\pm 0.012$	21.6
11	$35.9\pm 1.1$	$0.929\pm 0.012$	27.6

Table 3: Ablation on number of unrolling iterations

N

. Results at

R=4

Accelerated MRI Reconstruction

Accelerated MRI reconstruction recovers diagnostic images from undersampled k-space data to reduce scan time while preserving anatomical and quantitative information.

See all 85 papers

Physics-Driven 3D Gaussian Rendering for Zero-Shot MRI Super-Resolution
Mar 10, 2026Shuting Liu, Lei Zhang, Wei Huang +2Medical Image Super-Resolution Magnetic Resonance Imaging
School of Artificial Intelligence, Sichuan University, Chengdu 610065, China · Institute of High Performance Computing, A*STAR, Singapore 138632
MedGSSR: Generalizable Medical Image Super-Resolution 3D Reconstruction via Hierarchical Feed-forward Gaussian Splatting
Sep 6, 2026Chengkai Wang, Luoyu Hong, Yiting Zhao +7Medical Image Super-Resolution 3D Medical Imaging
University of British Columbia, Vancouver, Canada · Hangzhou Dianzi University, Hangzhou, China · ShanghaiTech University, Shanghai, China +1
Few-Shot Left Atrial Wall Segmentation in 3D LGE MRI via Meta-Learning
Mar 26, 2026Yusri Al-Sanaani, Rebecca Thornhill, Pablo Nery +5Medical Image Segmentation 3D Medical Image Segmentation
Department of Systems and Computer Engineering, Carleton University, Ottawa, Canada · Department of Radiology, Radiation Oncology, and Medical Physics, University of Ottawa, Ottawa, Canada · Division of Cardiology, Department of Medicine, University of Ottawa Heart Institute, Ottawa, Canada

Mar 10, 2026J/K move · Enter open · S save

Physics-Driven 3D Gaussian Rendering for Zero-Shot MRI Super-Resolution

Shuting Liu, Lei Zhang, Wei Huang, Zhao Zhang+1

School of Artificial Intelligence, Sichuan University, Chengdu 610065, China · Institute of High Performance Computing, A*STAR, Singapore 138632

Medical Image Super-Resolution Magnetic Resonance Imaging Physics-Guided Image Restoration 3D Gaussian Splatting

Open paper AskPDF arXiv

High-resolution Magnetic Resonance Imaging (MRI) is vital for clinical diagnosis but limited by long acquisition times and motion artifacts. Super-resolution (SR) reconstructs low-resolution scans into high-resolution images, yet existing methods are mutually constrained: paired-data methods achieve efficiency only by relying on costly aligned datasets, while implicit neural representation approaches avoid such data needs at the expense of heavy computation. We propose a zero-shot MRI SR framework using explicit Gaussian representation to balance data requirements and efficiency. MRI-tailored Gaussian parameters embed tissue physical properties, reducing learnable parameters while preserving MR signal fidelity. A physics-grounded volume rendering strategy models MRI signal formation via normalized Gaussian aggregation. Additionally, a brick-based order-independent rasterization scheme enables highly parallel 3D computation, lowering training and inference costs. Experiments on two public MRI datasets show superior reconstruction quality and efficiency, demonstrating the method's potential for clinical MRI SR.

Abstract

Accelerated MRI Reconstruction

Explore similar work

Physics-Driven 3D Gaussian Rendering for Zero-Shot MRI Super-Resolution

MedGSSR: Generalizable Medical Image Super-Resolution 3D Reconstruction via Hierarchical Feed-forward Gaussian Splatting

Few-Shot Left Atrial Wall Segmentation in 3D LGE MRI via Meta-Learning