Fast group matching for MR fingerprinting reconstruction

Area of Science:

• Quantitative imaging
• Biomedical engineering
• Medical physics

Background:

• Magnetic Resonance Fingerprinting (MRF) enables quantitative tissue mapping.
• MRF reconstruction involves computationally intensive matching of acquired data to Bloch simulations.
• Estimating T1, T2, proton density, and B0 requires efficient algorithms.

Purpose of the Study:

• Introduce a fast group matching algorithm (GRM) for MRF.
• Reduce the computational demand of MRF reconstruction.
• Enable clinically relevant reconstruction times and accuracy.

Main Methods:

• Developed a GRM algorithm exploiting correlations in MRF dictionaries.
• Utilized group-specific signatures for initial matching refinement.
• Employed group principal component analysis (PCA) for tissue type evaluation.
• Validated the approach using in vivo 3 Tesla brain data.

Main Results:

• GRM achieved MR parameter mapping within 2 seconds for a large MRF dictionary.
• Demonstrated an order of magnitude speedup over global PCA and two orders over direct matching.
• Maintained comparable accuracy with 1-2% relative error.

Conclusions:

• GRM is an efficient model reduction technique for MRF matching.
• The method enables clinically relevant reconstruction accuracy and speed.
• GRM is suitable for standard vendor computational resources.