Bayesian optimization of gradient trajectory for parallel-transmit pulse design.

Area of Science:

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

Background:

• Parallel-transmit MRI utilizes spoke pulses to enhance excitation homogeneity.
• Optimizing spoke trajectories is crucial for improving image quality and efficiency.
• Existing methods may not achieve global optima efficiently.

Purpose of the Study:

• To introduce an efficient global optimization algorithm, Bayesian Optimization of Gradient Trajectory (BOGAT), for parallel-transmit MRI.
• To optimize spoke pulse trajectories for single-slice and simultaneous multislice imaging.
• To improve excitation homogeneity and reduce energy deposition in MRI.

Main Methods:

• BOGAT employs an outer loop to optimize k_T-space positions, with RF coefficients optimized for each position.
• Bayesian optimization progressively estimates the cost function, automatically selecting k_T-space positions for efficient convergence.
• The algorithm was tested using a database of field maps from 85 MRI scans and prospectively in phantom and in vivo experiments.

Main Results:

• BOGAT converged within 10% of the global minimum cost in 93% of tested slices within 30 iterations.
• Compared to the vendor-provided Fourier transform approach, BOGAT achieved up to 56% lower flip angle RMS error (RMSE) or 55% lower pulse energy in phantoms.
• In vivo results showed up to 30% lower RMSE and 29% lower energy with minimal extra computation time (7.8s).

Conclusions:

• BOGAT efficiently estimates near-global optimum spoke positions for MRI pulse design.
• The algorithm significantly reduces flip-angle RMSE and pulse energy.
• The computation time is suitable for online optimization in parallel-transmit MRI applications.