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

2026-02-24

AI Summary

This study addresses the challenge of severe through-plane anisotropy in multi-stack thick-slice MRI, which hinders high-precision 3D analysis. To overcome this limitation, the work introduces 3D Gaussian Splatting into MRI reconstruction for the first time, proposing physically consistent magnetic Gaussian primitives to accurately model MR signal characteristics. The approach further incorporates a neural residual field to enhance fine structural details and employs a multi-resolution progressive training strategy to improve optimization stability and efficiency. Evaluated on the FeTA dataset, the method achieves a peak signal-to-noise ratio (PSNR) of 40.31 dB and demonstrates a 14-fold acceleration in reconstruction speed compared to existing techniques, delivering high-quality isotropic volume reconstructions with significantly improved computational efficiency.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Magnetic Resonance Imaging (MRI) is a crucial non-invasive imaging modality. In routine clinical practice, multi-stack thick-slice acquisitions are widely used to reduce scan time and motion sensitivity, particularly in challenging scenarios such as fetal brain imaging. However, the resulting severe through-plane anisotropy compromises volumetric analysis and downstream quantitative assessment, necessitating robust reconstruction of isotropic high-resolution volumes. Implicit neural representation methods, while achieving high quality, suffer from computational inefficiency due to complex network structures. We present M-Gaussian, adapting 3D Gaussian Splatting to MRI reconstruction. Our contributions include: (1) Magnetic Gaussian primitives with physics-consistent volumetric rendering, (2) neural residual field for high-frequency detail refinement, and (3) multi-resolution progressive training. Our method achieves an optimal balance between quality and speed. On the FeTA dataset, M-Gaussian achieves 40.31 dB PSNR while being 14 times faster, representing the first successful adaptation of 3D Gaussian Splatting to multi-stack MRI reconstruction.

Problem

• multi-stack MRI
• through-plane anisotropy
• isotropic reconstruction
• volumetric analysis
• quantitative assessment

Innovation

• Magnetic Gaussian
• 3D Gaussian Splatting
• multi-stack MRI reconstruction
• neural residual field
• progressive training