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

2026-03-10

AI Summary

This work proposes a zero-shot super-resolution framework for high-resolution MRI reconstruction that addresses the limitations of prolonged scan times, motion artifacts, and the trade-off between data dependency and computational efficiency in existing methods. By modeling tissue structures with an explicit 3D Gaussian representation embedded with physical priors, the approach integrates a physics-driven normalized Gaussian volume rendering mechanism and a tile-based, order-independent rasterization strategy to enable highly parallelized reconstruction. Notably, the method requires no paired training data and achieves state-of-the-art performance on two public MRI datasets, delivering significant improvements in both image quality and computational efficiency while substantially reducing training and inference costs.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Machine Learning: Matrix & Tensor Methods
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web

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.

Problem

• MRI super-resolution
• zero-shot learning
• high-resolution imaging
• motion artifacts
• paired-data limitation

Innovation

• Physics-Driven Rendering
• 3D Gaussian Representation
• Zero-Shot Super-Resolution
• MRI Signal Modeling
• Order-Independent Rasterization