Low rank approximation methods for MR fingerprinting with large scale dictionaries.

Area of Science:

• Magnetic Resonance Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Magnetic Resonance Fingerprinting (MRF) enables rapid quantitative MRI.
• Large-scale MRF applications face memory and computational challenges.
• Efficient low-rank approximation is crucial for MRF scalability.

Purpose of the Study:

• To develop memory-efficient low-rank approximation methods for large-scale MRF.
• To reduce the computational burden of MRF dictionary calculations.
• To enhance the clinical applicability of MRF.

Main Methods:

• Introduced a compressed MRF approach using randomized singular value decomposition.
• Exploited MRF dictionary structures within the randomized SVD space.
• Fitted dictionary structures to low-degree polynomials for high-resolution parameter mapping.

Main Results:

• Achieved significant memory savings, up to 1000x for one sequence and 15x for another.
• Generated T1, T2, and off-resonance maps in agreement with standard MRF.
• Validated the approach using in vivo 1.5T and 3T brain scan data.

Conclusions:

• Proposed methods are memory-efficient low-rank approximations for MRF.
• These techniques can benefit clinical MRF implementation.
• The methods show potential for large-scale MRF problems and high-resolution parameter spaces.