Automated design of pulse sequences for magnetic resonance fingerprinting using physics-inspired optimization

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Quantitative Imaging
• Medical Physics

Background:

• Magnetic Resonance Fingerprinting (MRF) enables quantitative tissue property extraction (T1 and T2 relaxation rates) using standard MRI hardware.
• MRF sequences offer numerous tunable parameters for optimizing precision and scan time.

Purpose of the Study:

• To develop a de novo automated design for MRF pulse sequences using physics-inspired optimization heuristics.
• To address systematic errors, particularly those from undersampling and phase variation, which dominate over random errors in clinical settings.

Main Methods:

• Employed physics-inspired optimization heuristics for automated MRF pulse sequence design.
• Utilized a cost function simulating systematic errors from Fourier undersampling and phase variation, differing from prior statistical error models.
• Experimental validation of the designed sequences.

Main Results:

• Achieved fourfold shorter scan times compared to previous human-designed MRF sequences with equivalent precision in T1 and T2.
• Discovered MRF pulse sequences with inherent robustness against shading artifacts caused by phase variations.
• The optimized sequences exhibited features qualitatively distinct from existing MRF sequences.

Conclusions:

• Automated design using first-principles simulation of systematic errors is effective for optimizing MRF sequences.
• The developed method significantly enhances scan efficiency and robustness in quantitative MRI.
• This approach paves the way for more practical and reliable MRF applications in clinical settings.