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

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Medical Imaging AI · Depth: Expert, extended

Summary

PhyMRI-SR introduces a physics-aware approach to MRI image super-resolution, challenging the conventional view of SR as a fixed-mapping problem. The method rethinks SR as a reconstruction task that accounts for the inherent resolution-SNR trade-off in MRI acquisition, proposing that optimal resolution is not always the highest. It adapts 2D Gaussian Splatting for resolution-agnostic rendering and incorporates three innovations: a prior-aware Gaussian representation using anatomical and imaging system priors, physics-constrained signal modeling to predict tissue parameters (ρ, R2) for biophysically plausible contrast, and a meta-learning framework to address data scarcity. Experiments on dynamic-resolution datasets (simulated 64mT-3T, real 3T-5T) and the FastMRI benchmark demonstrate state-of-the-art performance, with optimal visual quality often achieved at intermediate resolutions like x0.7 or x0.76. For instance, it achieved 34.26 dB PSNR and 0.962 SSIM at 4x scale on FastMRI.

Key takeaway

For AI Scientists developing MRI super-resolution models, you should rethink fixed-scale upsampling and consider the resolution-SNR trade-off. Your models can achieve superior diagnostic utility by dynamically optimizing for an intermediate resolution, rather than always targeting the highest. Implement physics-constrained signal modeling and leverage meta-learning with simulated data to improve generalization and biophysical plausibility on scarce real-world MRI datasets.

Key insights

MRI super-resolution benefits from physics-aware reconstruction that optimizes resolution-SNR trade-offs, not just upscaling fixed inputs.

Principles

• MRI resolution and SNR are inherently coupled.
• Optimal MRI SR resolution is often intermediate.
• Integrate anatomical and system priors for fidelity.

Method

PhyMRI-SR adapts 2D Gaussian Splatting for resolution-agnostic rendering, using prior-aware Gaussian representation, physics-constrained signal modeling predicting ρ and R2, and a meta-learning framework for domain adaptation.

In practice

• Use 2D Gaussian Splatting for continuous SR.
• Predict tissue parameters for biophysically plausible MRI.
• Employ meta-learning for scarce real-world MRI data.

Topics

• MRI Super-Resolution
• 2D Gaussian Splatting
• Physics-Aware Reconstruction
• Resolution-SNR Trade-off
• Meta-Learning
• Medical Imaging

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• PhyMRI-SR: Toward Physics-Aware MRI Image Super-Resolution — MRI super-resolution can be rethought as a physics-aware reconstruction problem, optimizing resolution-SNR configurations for dynamic inputs.
• CAHAL: Clinically Applicable resolution enHAncement for Low-resolution MRI scans — CAHAL enhances clinical MRI resolution by using physics-informed, native-space processing and a bivariate Mixture of Experts to prevent anatomical distortions.
• Physics-Driven Zero-Shot MRI Reconstruction with Non-local Image Priors — Combining physics-driven constraints with non-local image priors enhances zero-shot MRI reconstruction stability and performance.
• Recover Semantics First, Generate Better: Improved Latent Modeling for 3D MRI Reconstruction and Cross-Contrast Synthesis — Prioritizing semantic recovery in 3D MRI latent space compression significantly improves reconstruction fidelity and cross-contrast synthesis.
• Dual-Prior Guided Null-Space Learning with Mixture-of-Splines for Arbitrary Medical Slice Super-Resolution — DP-NSL reformulates super-resolution as a constrained recovery, using dual priors for measurement consistency and geometric continuity.
• Reduced NEXI protocol for the quantification of human gray matter microstructure on the Connectome 2.0 scanner — Explainable AI (XAI) can optimize complex MRI protocols, reducing scan time without losing data fidelity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.