ACCELERATING QUANTITATIVE MRI USING SUBSPACE MULTISCALE ENERGY MODEL (SS-MUSE)

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Multi-contrast MRI is crucial for tissue differentiation and quantitative mapping.
• Long acquisition times in 3D MRI limit isotropic resolution.
• Deep learning methods face challenges with large 3D datasets.

Purpose of the Study:

• To accelerate 3D multi-contrast MRI acquisition.
• To overcome limitations of deep learning for large-scale 3D volumes.
• To enable high-resolution 3D imaging with multiple contrasts.

Main Methods:

• Generalization of the plug-and-play multi-scale energy-based model (MuSE) to a regularized subspace recovery framework.
• Joint regularization of 3D multi-contrast spatial factors within a subspace formulation.
• Application of variable splitting optimization for efficient image recovery.

Main Results:

• Demonstrated computational efficiency in recovering 3D multi-contrast MRI data.
• Enabled faster acquisition of high-resolution 3D MRI scans.
• Provided a viable alternative to deep learning for accelerating 3D MRI.

Conclusions:

• The generalized MuSE model effectively accelerates 3D multi-contrast MRI acquisition.
• This approach addresses the computational and memory challenges of deep learning in 3D MRI.
• The method facilitates high-quality, isotropic 3D MRI with multiple contrasts.