A Minibatch Alternating Projections Algorithm for Robust and Efficient Magnitude Least-Squares RF Pulse Design in MRI

Area of Science:

• Magnetic Resonance Imaging
• Pulse Sequence Design
• Computational Electromagnetics

Background:

• Designing radiofrequency (RF) pulses for Magnetic Resonance Imaging (MRI) is crucial for image quality.
• Conventional methods can struggle with local minima and computational efficiency, especially at ultra-high field strengths.
• Inhomogeneities in the radiofrequency field require robust pulse design strategies.

Purpose of the Study:

• To develop a novel RF pulse design algorithm that overcomes limitations of existing methods.
• To improve the efficiency and reduce the cost of RF pulse design.
• To enhance the robustness of RF pulse design for ultra-high field MRI applications.

Main Methods:

• A magnitude-least-squares algorithm employing interleaved exact and stochastic updates.
• Utilizing small, randomly selected minibatches of measurements for inexact updates.
• Perturbing alternating projections to escape local minima and find optimal solutions.
• Investigating applications in RF shimming, parallel transmit spokes, and spectral-spatial RF pulse design.

Main Results:

• The algorithm consistently produced lower power and lower Root Mean Square Error (RMSE) solutions.
• Optimal minibatch size was characterized across various imaging parameters and hardware configurations.
• In vivo validation at 7 Tesla demonstrated improvements in image quality for RF-shimmed sequences.
• The method showed increased computational efficiency compared to conventional approaches.

Conclusions:

• The developed algorithm offers a more robust and computationally efficient method for designing RF pulses.
• It effectively corrects for magnetic field inhomogeneities at ultra-high field strengths.
• The approach leads to improved MRI image quality and lower energy deposition.