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

Shuting Liu; Lei Zhang; Wei Huang; Zhao Zhang; Zizhou Wang

cs.CVMar 10, 2026

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

Authors: Shuting Liu, Lei Zhang, Wei Huang, Zhao Zhang, Zizhou Wang

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

Abstract

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.

Figures & tables

Figure 1: The overall workflow of our method. 3D Gaussians are initialized solely from the coordinates of the low-resolution volume. During rendering, 3DGS relies on spherical harmonics (SH)-based view-dependent $\alpha$ -blending to produce RGB color, whereas our approach uses physics-grounded volume rendering to render voxel intensity. In the spatial distribution stage, 3DGS requires depth sorting followed by 2D splatting, while our method aggregates Gaussian contributions and performs brick-based rendering directly in 3D space to generate high-resolution MRI.

Figure 2: Illustration of the key MRI signal formation processes that challenge the original 3DGS framework and motivate our MRI-tailored Gaussian model. In MRI, protons first align with and precess in the static magnetic field $B_{0}$ (a), then an RF pulse tips them into the transverse plane where they dephase and relax (b), producing the time‐dependent signal used to reconstruct the image.

Method	2 $\times$	3 $\times$	4 $\times$	Arbitrary
Conventional methods
Bicubic	33.75, 0.947	30.74, 0.916	28.67, 0.872	31.56, 0.933
Supervised methods
MIASSR	37.56 , 0.966	32.80, 0.898	30.18, 0.807	N/A
ARSSR	35.58, 0.977	34.03 , 0.959	30.63 , 0.938	N/A
Zero-shot methods

Table 1: PSNR/SSIM for multi-scale SR on the MSD dataset [ 18 ] . Bold and underline indicate the best and second best results.

Conventional methods
Method	2 $\times$	3 $\times$	4 $\times$
Bicubic	37.29, 0.969	33.21, 0.913	29.62, 0.863
Supervised methods
MIASSR	38.71, 0.974	33.39, 0.921	30.10, 0.835
ARSSR	40.99 , 0.995	36.87 , 0.988	34.45 , 0.979
Zero-shot methods

Table 2: PSNR/SSIM for multi-scale SR on the FeTA dataset [ 21 ] .

Figure 3: Visual comparison of the proposed method against state‑of‑the‑art approaches on two public datasets. Left column: Full reconstructed images. Top-right: Cropped region (indicated by a green box on the ground truth). Bottom-right: Pixel-wise error maps between reconstructed patches and corresponding HR reference patches.

$a$	rt	2 $\times$	3 $\times$	4 $\times$
$\times$	$\times$	17.99, 0.825	19.42, 0.817	19.76, 0.790
$\times$	$\checkmark$	34.56, 0.955	29.83, 0.918	27.77, 0.901
$\checkmark$	$\checkmark$	43.03 , 0.988	38.57 , 0.972	36.15 , 0.957

Table 3: Ablation on amplitude (

A

) and relaxation proxy (T) for 3D MISR on the MSD dataset [ 18 ] .

Figure 4: Radar chart of zero‑shot MRI SR methods (NeRF, CuNeRF, Ours) on MSD [ 18 ] at 2× (left) and 4× (right) upsampling. Metrics: PSNR/SSIM for quality and training time, inference time, peak VRAM for resources.

Explore similar work

PhyMRI-SR: Toward Physics-Aware MRI Image Super-Resolution
Jul 7, 2026Lihua Wei, Huatong Gao, Jia Gong +4Medical Image Super-Resolution Magnetic Resonance Imaging
School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210, China · School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210, China, and also with the Shanghai Academy of AI for Science, Shanghai, China · Shanghai Academy of AI for Science, Shanghai, China +4
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
M-Gaussian: An Magnetic Gaussian Framework for Efficient Multi-Stack MRI Reconstruction
Feb 24, 2026Kangyuan Zheng, Xuan Cai, Jiangqi Wang +6Magnetic Resonance Imaging 3D Gaussian Splatting
School of Biomedical Engineering, Shenzhen Campus of Sun Yat-sen University, Shenzhen, 518107, China. · the School of Information Science and Engineering, Shandong Normal University, Shandong, China. · School of Computing and Data Science, The University of Hong Kong, Hong Kong, China.

Jul 7, 2026J/K move · Enter open · S save

PhyMRI-SR: Toward Physics-Aware MRI Image Super-Resolution

Lihua Wei, Huatong Gao, Jia Gong, Zhiyu Tan+3

School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210, China · School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210, China, and also with the Shanghai Academy of AI for Science, Shanghai, China · Shanghai Academy of AI for Science, Shanghai, China+4

Medical Image Super-Resolution Magnetic Resonance Imaging Physics-Informed ML Gaussian Splatting

Open paper AskPDF arXiv

Magnetic resonance imaging (MRI) super-resolution is vital for improving diagnostic accessibility, yet most methods treat it as a deterministic mapping from a fixed low-resolution input to a high-resolution target. This overlooks a key property of MRI acquisition physics: spatial resolution and signal-to-noise ratio (SNR) are inherently coupled, making any given low-resolution scan merely one of many possible realizations under varying acquisition trade-offs. We rethink MRI super-resolution as a physics-aware reconstruction problem, in which the goal is to identify the optimal resolution-SNR configuration and then super-resolve it to obtain high-quality MRI results. A key implication of this formulation is that MRI resolution becomes dynamic rather than fixed. To handle such resolution-heterogeneous inputs, we adapt 2D Gaussian Splatting (2D GS) to MRI by formulating reconstruction as a coordinate-based, resolution-agnostic rendering problem. To further enhance fidelity, we introduce three innovations: (1) a prior-aware Gaussian representation that combines an Anatomical Structure Prior for tissue-specific kernel initialization with an Imaging System Prior that captures hardware characteristics via a covariance dictionary; (2) a physics-constrained signal modeling scheme that predicts intrinsic tissue parameters (proton density rho and effective relaxation rate R2) and synthesizes intensities through governing physical equations, ensuring biophysically plausible contrast; and (3) a meta-learning framework that alleviates paired-data scarcity by pretraining on simulated data and adapting to real-world conditions. Extensive experiments on dynamic-resolution datasets and standard benchmarks demonstrate that our method achieves state-of-the-art performance, highlighting its strong potential for clinical deployment.

Abstract

Explore similar work

PhyMRI-SR: Toward Physics-Aware MRI Image Super-Resolution

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

M-Gaussian: An Magnetic Gaussian Framework for Efficient Multi-Stack MRI Reconstruction